Determinants of Mobile Advertising Effectiveness: A Field Experiment

_______________ Yakov BART Andrew T. STEPHEN Miklos SARVARY 2012/38/MKT

Electronic copy available at: http://ssrn.com/abstract=2029496

Determinants of Mobile Advertising Effectiveness: A Field Experiment

Yakov Bart* Andrew T. Stephen** Miklos Sarvary***

March 21 2012

* Yakov Bart (Yakov.Bart@insead.edu) is an assistant professor of marketing at INSEAD, 1 Ayer Rajah Avenue, Singapore 138676. Andrew T. Stephen (AStephen@katz.pitt.edu) is an assistant professor of business administration and Katz Fellow in marketing at the Joseph M. Katz Graduate School of Business, University of Pittsburgh, 318 Mervis Hall, Pittsburgh, PA 15260. Miklos Sarvary (Miklos.Sarvary@insead.edu) is Professor of Marketing and The GlaxoSmithKline Professor of Corporate Innovation at INSEAD, Boulevard de Constance, Fontainebleau 77305, France. All authors contributed equally to this work and their names are listed in random order. This research was funded by a Google-WPP Marketing Research Award and the INSEAD Alumni Fund. The authors thank George Pappachen, Ali Rana, Kara Manatt, and Aaron Katz for their assistance and support with this research. This paper can be downloaded without charge from the Social Science Research Network electronic library at: http://ssrn.com/abstract=2029496

*

Assistant Professor of Marketing at INSEAD, 1 Ayer Rajah Avenue 138676 Singapore. Email Yakov.bart@insead.edu Assistant Professor of Business Administration & Katz Fellow in Marketing University of Pittsburgh, Joseph M. Katz Graduate School of Business318 Mervis Hall, Pittsburgh, PA 15260, USA. Email: AStephen@katz.pitt.edu

**

***

Professor of Marketing, The GlaxoSmithKline Chaired Professor of Corporate Innovation at INSEAD Boulevard de Constance, Fontainebleau 77305, France. Email : Miklos.Sarvary@insead.edu

A Working Paper is the author’s intellectual property. It is intended as a means to promote research to interested readers. Its content should not be copied or hosted on any server without written permission from publications.fb@insead.edu Click here to access the INSEAD Working Paper collection

Electronic copy available at: http://ssrn.com/abstract=2029496

ABSTRACT

Mobile advertising is one of the fastest growing advertising formats, with U.S. spending in 2011 estimated at $1.2 billion and global spending forecasted to reach over $20 billion by 2015. Interestingly, despite the rapid penetration of sophisticated handsets a growing proportion of mobile advertising spending consists of display advertising, which is known for its limited capacity for the transfer of information. This paper examines why and under what conditions such ―low-fidelity‖ mobile display advertising is effective in increasing consumers’ purchase intentions. We draw on consumer psychology to identify these conditions and verify our hypotheses in a field experiment involving 54 national mobile display advertising campaigns that ran between 2007 and 2010 and involved 27,753 participants. Our results indicate that low-fidelity mobile advertising campaigns are effective when they are for products that trigger further thought and consideration, which includes campaigns for high (versus low) involvement products, and for products that are seen as more utilitarian (versus more hedonic).

2

Electronic copy available at: http://ssrn.com/abstract=2029496

1. INTRODUCTION

Mobile advertising has become one of the fastest-growing tools for reaching consumers over the last few years. Overall spending on mobile advertising has been growing quickly, shifting from $416 million in 2009 to $743 million in 2010 and $1.2 billion in 2011, and is forecasted to reach $4.4 billion by 2015 in the U.S. market alone (eMarketer.com 2011). Globally, mobile advertising spending is expected to reach $20.6 billion by 2015 (Gartner 2011). While only 37% of U.S. marketers surveyed in May 2011 used mobile advertising as one of their communications channels, 58% said that they intended to use mobile advertising in the near future. The popularity of the mobile channel as a way for marketers to reach consumers is expected to continue to increase, fueled by the proliferation of mobile connected devices throughout the world. Thus, there is an increasingly high potential reach for mobile advertising. Mobile advertising comes in a variety of forms, including display, SMS/text message, locationbased, and rich media. Display advertising is the most pervasive, largely because advertisements of this kind are compatible with almost all types of mobile phones currently in use around the world (e.g., smartphones such as iPhone, older smartphones such as Blackberry, and lower-cost feature phones such as many produced by Nokia and Motorola). Of these, mobile banner images displayed on top of the screen in a Wireless Application Protocol (WAP) browser constitute the most common form of such advertising. Recently, this advertising unit has also been embedded in various mobile smartphone applications (apps), which has afforded app developers with another potential source of revenue. Figure 1 shows two examples of mobile display advertisements and their placements in WAP browsers (panel A) and apps (panel B). Interestingly, while the popularity of some other forms of mobile advertising has either exhibited no growth or started to decline (e.g., SMS advertisements), the popularity of mobile display advertising has been steadily increasing. Specifically, the share of total U.S. mobile advertising held by display is expected to grow from 35% in 2011 to 45% by 2015 (eMarketer.com 2011). [INSERT FIGURE 1 ABOUT HERE] The fast growth of mobile display advertising is somewhat surprising because this advertising format has severe limitations in its ability to engage consumers. Mobile display advertisements occupy a very small portion of an already small screen, can only contain little amounts of information, and consumers are typically ―on the move‖ and therefore exposed to a lot of other stimuli. These factors conceivably can severely constrain the effectiveness of mobile display advertising, casting doubt on its ability to have an impact on consumers’ attitudes and behaviors. However, industry studies generally indicate that mobile advertising can be very effective (Pappachen and Manatt 2008). Given the evidence from practice and the inherent limitations of this form of digital advertising, an important question is when should advertisers use mobile display advertising? Existing research

3

provides little guidance. Formal studies on mobile advertising are rare and have severe limitations both in terms of their methodologies and their scope. In particular, previous studies of mobile advertising effectiveness (Barwise and Strong 2002; Drossos et. al. 2007; Tsang, Ho and Liang 2004) have focused only on SMS advertising, a format that is declining in share. While these studies all show a positive effect of SMS advertisements on various stages of consumer decision-making, each set of findings comes from a very limited sample of participants. Moreover, these studies are primarily motivated by pragmatic considerations and therefore provide little insight into the psychological mechanisms that drive mobile advertising effectiveness. For these reasons, existing empirical evidence provides insufficient guidance on the conditions under which mobile advertising is likely to be effective. Given the dominant share and rising popularity of mobile display advertising, it is important to develop a better understanding of when this type of mobile advertising should be utilized and why it is likely to be effective under some but not all conditions. Put simply, for which types of products, services, or brands1 is mobile display advertising most likely to be effective? We address this fundamental question in this paper. To the best of our knowledge, this research is the first extensive and thorough empirical study in a field setting to address this question. Our approach proceeds in three steps. We start by introducing a new typology of digital advertising that accommodates technology restrictions. Technology restrictions are particularly relevant to mobile advertising since low bandwidth (Internet speed) and restricted physical space (screen size) are major limitations faced by advertisers that constrain their options with respect to advertising copy, visual richness, and information content. Our typology classifies digital advertising into two types: high-fidelity and low-fidelity. We argue that mobile display advertisements almost always fall into the latter category, which in practice means that advertising messages carry relatively little information and often are low quality, particularly visually. It is therefore somewhat surprising that prior research (including industry studies; e.g., Pappachen and Manatt 2008) finds mobile advertising to be generally very effective. Next, we develop a psychological theory of consumer response to low-fidelity advertising (such as mobile display) based on the classic elaboration likelihood model of persuasion in communication (e.g., Petty et. al. 1983). We propose a number of hypotheses that collectively define conditions under which lowfidelity messages are expected to be persuasive. Finally, we test specific hypotheses derived from this theory in a large-scale field experiment involving 54 different mobile advertising campaigns conducted between mid 2007 and mid 2010 that covered 13 diverse industries and involved 27,753 consumers. The empirical findings generally support our theory and hypotheses. To preview our results, we find positive advertising treatment effects on consumer purchase intent only for products that are utilitarian (as opposed to hedonic) and for which product consideration and purchase decision-making
1

Hereafter we refer to products, services, or brands simply as ―products.‖

4

processes are high involvement (as opposed to low involvement). In a number of other conditions, however, we find either no treatment effects for mobile display advertising or negative effects (i.e., when consumers exposed to an advertisement become less inclined to purchase that product). Thus, the current research represents an important step toward helping scholars and managers better understand when lowfidelity advertising, and, in particular, mobile display advertising, is appropriate, and which types of products are best suited for being advertised through this medium.

2. RELATED LITERATURE

Our work is related to two broad literature streams. The first stream broadly covers advertising effectiveness in the context of digital media. Beyond the limited literature on SMS advertising mentioned earlier, the broader context of digital advertising has gained attention recently among both information systems and marketing scholars. The literature has examined various aspects associated with digital advertising effectiveness and, more generally, effects of media activity on business outcomes such as sales. For example, Goldfarb and Tucker (2011) study determinants of online display advertising effectiveness, Ghose and Yang (2009) examine a similar issue in the domain of online search engine advertising, and Danaher, Lee, and Kerbache (2009) consider media mix issues in digital advertising. In the context of so-called ―earned‖ digital media, recent work has compared the sales response to mentions of a brand in social media to traditional media publicity (Stephen and Galak 2012), explored how online word-of-mouth affects the growth of websites and social networking platforms (Trusov, Bucklin, and Pauwels 2009), developed economic models of users’ motivations to contribute content on social media platforms such as Twitter (Toubia and Stephen 2012), and analyzed the sales impact of online usergenerated product reviews (Dellarocas, Zhang, and Awad 2007). Although studies of mobile media usage and the mobile Internet have also begun to emerge (e.g., Garg and Telang 2011; Ghose and Han 2011; Shim, Park, and Shim 2008), mobile advertising has not received much attention from researchers. In particular, prior research on mobile advertising effectiveness has not considered display advertising and has not examined its effectiveness within the context of a large-scale field experiment and across multiple industries and product categories. Further, prior research has tended to show only cases in which mobile advertisements are effective. While it is possible that mobile advertisements are generally effective in persuading and influencing consumers’ attitudes and behaviors, it is unlikely that their effects are universally positive. Thus, a more general and comprehensive understanding of this increasingly important and popular advertising medium is needed. The second stream of research is grounded in consumer psychology. In particular, we heavily build on a classic model of persuasion in communication: the elaboration likelihood model (ELM; Petty

5

and Cacioppo 1981, 1983) and its application to advertising (e.g., Petty, Cacioppo, and Schumann 1983; Petty and Cacioppo 1986). We discuss ELM and related concepts in more detail when we develop our hypotheses below. In addition to the literature on communication persuasion, we also draw on the consumer psychology literature on emotions (e.g., Avnet, Pham, and Stephen 2012; Pham, Lee, and Stephen 2012) and consumption of hedonic versus utilitarian goods (Dhar and Wertenbroch 2000).

3. INDUSTRY CONTEXT: TYPOLOGY OF DIGITAL ADVERTISING MEDIA

We introduce a typology of digital advertising media in Table 1 in which media are classified as either high- or low-fidelity. We borrow the concept of ―fidelity‖ from literature in computer science and engineering where fidelity refers to a property of prototypes, websites, and images. For example, lowfidelity prototypes in engineering have generally limited functionality and are relatively low quality. Lowfidelity prototypes are often static and lack interactivity. A low-fidelity prototype of a website design could be as simple as a pencil-and-paper sketch of how a website should look and its basic functions (cf. Rudd, Stern, and Isensee 1996). In contrast, high-fidelity prototypes are much more thoroughly developed and ―fleshed out,‖ and therefore tend to be more realistic and detailed, feature a higher degree of functionality and interactivity, and are generally of better quality. Importantly, low-fidelity prototypes tend to convey limited information, whereas high-fidelity prototypes offer more extensive information. [INSERT TABLE 1 ABOUT HERE] Adapted to the context of digital advertising, low- versus high-fidelity refers to how much information can be feasibly conveyed through a particular medium, the medium’s level of quality in presenting that information, and aspects associated with the consumer experience such as the degree to which the medium is interactive. Examples of high-fidelity digital advertising media include rich media advertisements such as video-based interstitials or interactive display or banner advertisements, advertisements that are based on Flash or HTML5 technologies, and other types of media that are highly interactive and are able to present relevant information to consumers in a high-quality dynamic format. In contrast, low-fidelity digital advertising media include advertisements that are fully text based (e.g., Google Adwords), contain low-resolution and static images (e.g., basic website display banners), or a combination of text and images in a static, non-interactive, and size-constrained format (e.g., Facebook advertisements). In the domain of mobile advertising, the vast majority of options managers currently have at their disposal fall into the low-fidelity category. For example, SMS advertisements only contain text. MMS advertisements contain images but are still very basic. Advertisements on smartphones with higher-resolution screens can be high-fidelity, such as advertisements on Apple’s iOS devices published through the iAd platform (e.g., advertisements can feature video, high-quality images, and be interactive).

6

However, a more common mobile advertising unit, including on current-model smartphones, is the mobile display advertisement, which is another example of a low-fidelity advertisement. While high-fidelity digital advertisements are more frequently used on websites, they are less common on mobile devices due to technology restrictions. Even though many mobile devices have the ability to access the Internet via high-speed connections (e.g., Wi-Fi), often their connections rely on mobile Internet technologies such as 3G and 4G networks. These Internet connections have limited bandwidth and, therefore, limit what advertisers can realistically do with rich media, particularly with respect to streaming video and other dynamic interactive elements. Another important aspect that limits the widespread use of high-fidelity advertising in mobile channels is screen size. With the exception of tablet devices (e.g., iPad), most mobile connected devices have relatively small screens. This simply means that advertisers cannot fit as much content or information into an advertisement, even if it occupies the whole screen (which is rare, particularly with mobile display advertisements). Finally, some mobile device platforms and operating systems make rich-media and interactive advertisements difficult or impossible to implement. For example, while Flash has been a defacto standard for interactive advertisements on mobile websites, this software is not allowed on Apple’s iOS platform (i.e., iPhone and iPad). Moreover, given the heterogeneity in types of mobile devices and those devices’ technical capabilities, advertisers often prefer to use simpler, low-fidelity advertising media in mobile channels to ensure the compatibility of their advertisements with as many devices and networks as possible, increasing an advertisement’s potential reach. Taken together, these factors make it likely that the lowfidelity type of mobile advertising will remain highly popular in the foreseeable future.

4. THEORY AND HYPOTHESES

Despite their apparent inferiority, low-fidelity mobile advertisements of various types have been shown in prior research to be effective in influencing consumers’ attitudes and behaviors (e.g., Barwise and Strong 2002; Drossos et al. 2007; Pappachen and Manatt 2008). Although we argue below that lowfidelity mobile display advertisements should only be effective under specific conditions, previous findings imply that this type of advertising unit at least has the potential to be persuasive. Marketers and psychologists have studied advertising effectiveness and consumer responses to advertising for decades. In the psychology literature, a classic model of persuasion in communication is the elaboration likelihood model (ELM; Petty and Cacioppo 1981, 1983). The ELM is an informationprocessing model of how individuals’ attitudes are formed or changed in response to potentially persuasive stimuli such as advertisements. The ELM postulates two routes through which persuasion can

7

occur: the central route and the peripheral route.2 Persuasion through the central route implies consumers deliberately process message-relevant information. In contrast, in the case of the peripheral route attitudes are determined by environmental cues and not the message or information itself. Individuals process information in a more cognitive manner under the central route, which in the mobile display advertising case would mean that they pay attention to the content of the advertisement and thoughtfully consider it (i.e., with a high ―elaboration likelihood‖ and thus a high degree of cognitive elaboration). On the other hand, individuals rely more on affective evaluation of stimuli under the peripheral route. In fact, the literature on affect-as-information suggests that a person’s current affective state (i.e., mood) or their feelings toward a stimulus could themselves be peripheral cues (cf. Avnet et al. 2012; Pham et al. 2012; Schwarz, Bless, and Bohner 1991). In the case of mobile display advertising, peripheral-route processing translates into paying less direct attention to the content of the advertisement and relying more on other cues that are not contained within the advertisement itself and therefore less likely to be controlled by the advertiser (i.e., with a low ―elaboration likelihood‖ and thus a low degree of cognitive elaboration). By their very nature, low-fidelity mobile display advertisements lend themselves more to being processed through the peripheral route than through the central route. This, however, may be problematic. In a relatively low-information environment there is an increased risk that consumers will misunderstand an advertisement, not pay sufficient attention to it, or become influenced by cues that are tangential to the advertisement and the product being advertised. Put simply, consumers may draw unintended conclusions (if they draw any conclusions at all). Moreover, the peripheral cues that influence consumers under the peripheral route to persuasion may negatively affect their attitudes and intentions. For example, a lowfidelity advertisement for an airline when processed along the peripheral route could cue memories of ruined vacations due to delayed flights and lost baggage, which may result in a consumer’s attitude toward the advertised airline becoming less positive (or more negative). This would be less likely under the central route, since information contained in the advertisement itself (e.g., ―We have a 99% on-time departure record‖) would be thoughtfully processed and directly linked to the consumer’s attitude. Another example is low-fidelity advertisements for movies. Under peripheral-route processing a movie advertisement with low image quality may result in a consumer (unconsciously) expecting the movie itself to be of low quality, whereas such an association would be less likely under central-route processing. Given these risks, we only expect low-fidelity mobile advertisements to be effective in having a positive effect on consumer attitudes (e.g., purchase intent) when they are processed through the central

2

Related ―dual process‖ models in psychology postulate information processing and responding to stimuli as following either a more cognitive, systematic, and reason-based process versus a more affective, heuristic-based, and intuitive process (e.g., Chaiken 1980; Epstein and Pacini 1999).

8

route. Two determinants of which route an individual takes are processing elaboration ability and processing motivation (Petty and Cacioppo 1979, 1986). Processing ability refers to the relative ease or difficulty of elaborating on a stimulus when processing its information. This may be an individual difference, but can also be associated with product type. Processing ability can also be more generally associated with the presence or absence of distractions external to the stimulus or distortion associated with the processing of the stimulus itself. For example, the environmental cues that are attended to under peripheral-route processing but are not in fact central to the task at hand may be a source of distortion or distraction that lowers processing elaboration ability, thus lowering elaboration likelihood and making central-route processing less likely to occur (Petty, Ostrom and Brock 1981). Processing motivation refers to the extent to which a person is motivated to engage in effortful processing of a stimulus. When consumers are highly motivated they are expected to have higher elaboration likelihood and will process through the central route. The advertisement itself is unlikely to provide high processing ability or high processing motivation in the case of low-fidelity mobile display advertising. Certain characteristics of the product, however, may promote processing motivation and enhance processing ability, and thus potentially increase the likelihood the central route is taken, and thus the nature and extent of persuasion.

4.1 Processing Ability and Product Type We first consider a product characteristic associated with the consumer having a higher or lower processing ability: whether the product is more utilitarian or more hedonic (product type). Utilitarian products tend to be instrumental and functional, whereas hedonic products are associated with experiential consumption, pleasure, excitement, and fun (Dhar and Wertenbroch 2000; Holbrook and Hirchman 1982). Because of their functional nature, utilitarian products are more likely to be processed in a deliberative, thoughtful, and cognitive manner. Importantly, when thinking about utilitarian products, consumers are less likely to be distracted by thoughts not central to the evaluation of the product. Further, since utilitarian products tend to be more functional and used for achieving instrumental goals, consumers tend to find it easier to imagine consuming the product and deriving value from consumption. Thus, they are expected to have higher processing ability when considering utilitarian products. Hedonic products, on the other hand, lend themselves more to affective and heuristic-based processing. Further, they are often more experiential in nature and require more holistic judgments to evaluate. These aspects tend to make it more difficult to imagine oneself consuming and deriving value from hedonic products, and also make it harder to avoid judgments being distorted by peripheral cues. Thus, hedonic products are expected to be associated with lower processing ability. These arguments suggest that utilitarian (hedonic) products are associated more with the central (peripheral) route (Geuens, Pham, and De Pelsmaker 2010).

9

In an advertising context, a consumer presented with an advertisement for a utilitarian product has better ability to directly scrutinize the message’s arguments. This is harder when considering advertisements for hedonic products, since hedonic products lend themselves more to intuitive or ―gut feeling‖ assessments relying on peripheral cues and heuristics. Thus, compared to hedonic products, ability to process advertisements for utilitarian products is expected to be higher. Elaboration is therefore more likely, and consequently it is likely that utilitarian products will be processed through the central route and hedonic products will be processed through the peripheral route. Importantly, this also means that low-fidelity advertisements for utilitarian products are less susceptible to being misunderstood or otherwise erroneously processed. Thus, low-fidelity mobile display advertisements are more likely to have positive effect on consumers’ attitudes such as purchase intent for utilitarian products than for hedonic products. Stated formally: Hypothesis 1: After exposure to low-fidelity advertising, consumers’ increase in purchase intent for utilitarian products will be larger than for hedonic products.

4.2 Processing Motivation and Product Involvement Previous research on advertisement processing motivation shows that higher processing motivation increases the likelihood of central-route processing and increases the extent to which an advertisement impacts purchase intent (Mackenzie and Spreng 1992). In other words, when a consumer is exposed to an advertisement and is highly motivated to process it his/her resulting attitudes should be less affected by peripheral cues, s/he should be less likely to use heuristic-based processing, and s/he should be less influenced by feelings-based inferences. Psychologists have identified personal relevance as one of the most important variables affecting the motivation to process a persuasive message (Petty and Cacioppo 1979). High personal relevance means that a product is intrinsically important (Sherif and Hovland 1961) and is expected to have significant consequences for one’s life (Apsler and Sears 1968). In consumer behavior this is often referred to as a product being high consideration because high personal relevance and intrinsic importance implies that a consumer will engage in a high level of effortful processing when considering such a product or information associated with that product (e.g., an advertisement). Here we more generally refer to this as high involvement. High-involvement products (or advertisements for high-involvement products) therefore will generally be examined in greater detail and considered with greater care. Conversely, consumers will not focus on low-involvement products with as much detail or care. Put simply, processing motivation is expected to be higher (lower) for advertisements for products that are high (low) involvement. Consequently, it is likely that high-involvement products will be processed through the central route and

10

low-involvement products will be processed through the peripheral route. As with utilitarian products, this means that low-fidelity advertisements for high-involvement products will also be less susceptible to being misunderstood or otherwise erroneously processed. Low-fidelity mobile display advertisements are therefore more likely to have positive effect on consumers’ attitudes such as purchase intent when they are for high-involvement products than when they are for low-involvement products. Stated formally: Hypothesis 2: After exposure to low-fidelity advertising, consumers’ increase in purchase intent for high-involvement products will be larger than for low-involvement products.

4.3 Processing Ability and Processing Motivation As noted above, both processing ability and processing motivation affect how a stimulus is processed. Even if a consumer is highly motivated to process an advertisement (i.e., for a highinvolvement product), it may be difficult for them to do this—and take the central route to persuasion— unless they simultaneously have a high processing ability (i.e., the product type is utilitarian). Thus, we expect low-fidelity advertisements to be most persuasive when they are for products that promote high processing motivation and high processing ability. This implies a product type product involvement

interaction such that the advertising effect is highest when processing motivation and processing ability are high, which in our setting corresponds to a utilitarian, high-involvement product. Stated formally: Hypothesis 3: After exposure to low-fidelity advertising, consumers’ increase in purchase intent will be highest for utilitarian, high-involvement products (i.e., there will be a positive interaction between product type and involvement).

4.4 Recent Advertising Exposure The timing of the placement of an advertisement that is part of a larger multi-channel campaign3 may also affect processing and the persuasion outcome. This is related to whether a consumer recalls being recently exposed to advertising for a given product. The effect of recent advertising exposure on the process through which a subsequent advertisement is evaluated and the resultant effect on purchase intent, however, is unclear. For instance, consumers who have not recently been exposed to an advertisement for a product may have higher motivation to process an advertisement when they see it (for the first time) because it is novel. On the other hand, consumers may have lower processing ability because they are inherently less familiar with the product due to not having a prior, recent opportunity to consider and evaluate that product in response to an advertising exposure. These potential processes imply opposite
3

Mobile channels are rarely the only channel used in a campaign.

11

predictions for the effect of recent advertising exposure on the current advertising effect on purchase intent. However, following the ELM we expect processing motivation to be marginally more important than processing ability. That is, we should find that not having been recently exposed to an advertisement for a product will increase the impact of the current exposure on purchase intent. Stated formally: Hypothesis 4: After exposure to low-fidelity advertising, consumers’ increase in purchase intent will be larger when the consumer has not recently been exposed to advertisements for that product.

5. DATA AND METHODOLOGY

5.1 Mobile Advertising Field Experiment Data Our data are from a large digital market research company in the United States that specializes in measuring the effectiveness of digital advertising, including mobile display advertisements. Specifically, we examine 54 mobile display advertising campaigns for brands across 13 different industries (e.g., consumer packaged goods, entertainment, financial services, health care) that were run between July 2007 and June 2010. The company worked with mobile advertising networks and mobile service providers to place advertisers’ low-fidelity mobile advertisements on specific mobile (WAP) webpages on consumers’ mobile devices. Across the 54 campaigns, 27,753 individuals participated. Each campaign was run as a field experiment whereby participants were randomly assigned to one of two conditions: (i) exposed, where an advertisement was displayed on their screen when they browsed to a particular mobile webpage, or (ii) control, where they were not exposed to an advertisement when they browsed to that webpage. 4 In both conditions, the webpage also included a prominently placed banner (that was the same size as the advertisement in the exposed condition) that invited participants to complete a very short survey that measured purchase intent on a five-point scale (―Next time you are looking to purchase product category, how likely are you to purchase brand?‖ 1 = very unlikely, 5 = very likely). A second survey item measured whether or not participants had been recently (prior 30 days) exposed to advertising for the focal product (yes/no). An example of this arrangement is shown in Figure 2. In total, 14,457 (52.1%) participants were assigned to the exposed condition, and 13,296 (47.9%) participants were assigned to the control condition. [INSERT FIGURE 2 AND TABLE 2 ABOUT HERE]

4

This approach is identical to advertising effectiveness testing procedures used in other digital advertising media, such as webpage display advertising (e.g., Goldfarb and Tucker 2011).

12

Various summary statistics for the campaigns are reported in Table 2. The mean number of participants per campaign was 513.94 (SD = 554.87) and the mean campaign length, which was the number of days the advertisement in the exposed condition was displayed and the product-specific purchase intent measurement was undertaken, was 54.80 days (SD = 32.79). As mentioned above, a variety of industries were represented in this set of campaigns. The most represented industry was consumer packaged goods (33.33% of campaigns), followed by financial services (16.67%) and automotive (12.96%). The majority of the campaigns were for products (85.19%, vs. services), businessto-consumer brands (96.30%, vs. business-to-business), and for existing products or brand extensions (92.59%, vs. new products).

5.2 Data Analysis Methodology To test our hypotheses we first classified each campaign on the two theoretically relevant dimensions of product involvement (high vs. low) and product type (hedonic vs. utilitarian). The company provided us with campaign-level data for classifying the product in each campaign as either high or low involvement. For product type we examined each campaign’s product and judged it as either more hedonic or more utilitarian, following Dhar and Wertenbroch (2000).5 This resulted in a 2 (product involvement: high, low) 2 (product type: hedonic, utilitarian) campaign classification. We report the

distributions of campaigns and participants across these cells in Table 3, and the exposed versus control condition sizes for each cell in Table 4. Since the field experiment involved real advertising campaigns, we did not expect the distribution of campaigns and participants across these four cells to be uniform. [INSERT TABLES 3 AND 4 ABOUT HERE] We treated each participant as the unit of analysis. We estimated the average treatment effect for exposed versus control on purchase intent, specifically focusing on how both the sign and size of the effect varied according to (i) whether the participant had recent exposure to advertising for the focal product (yes, no), (ii) product involvement (high, low), and (iii) product type (hedonic, utilitarian). For this we estimated a random effects linear model to regress purchase intent on main effects for product type (utilitarian = 1, hedonic = -1), product involvement (high = 1, low = -1), recent exposure (yes = 1, no = -1), treatment condition (exposed = 1, control = -1), and all two-, three-, and four-way interactions. We also included fixed effects (dummy variables) for industry, year, whether the advertised product was new or existing/extension, whether the product was an actual product or a service, and whether the product

5

We also took into account how each product was framed or positioned in the campaign when making these judgments.

13

was business-to-consumer or business-to-business, as well as a covariate for campaign length (in days).6 To capture unobserved heterogeneity across campaigns we estimated a campaign random effect.

6. RESULTS

6.1 Hypothesis Tests Results from the random effects linear model are reported in Table 5, where we report estimates from two nested models. The first model is a full model including all effects, including those of the control variables (e.g., fixed effects for industry and year). The second model estimated only the main and interaction effects for the four variables of interest and not the control variables’ effects. Although the fit of the second model is slightly better, we base our findings on the full model. In any case, the effects required for testing our hypotheses did not differ in terms of sign, size, or significance. [INSERT TABLES 5 AND 6 ABOUT HERE] We now turn our attention to testing our hypotheses by focusing on estimated means for purchase intent and treatment effects (differences in estimated means between exposed and control conditions). A significant positive (negative) treatment effect indicates that the average purchase intent was significantly higher (lower) for exposed participants than control participants. Table 6 lists estimated means and treatment effects broken down by product type and involvement. Hypothesis 1 predicted a larger positive treatment effect for purchase intent for utilitarian products than for hedonic products. This implies a significant positive utilitarian exposed interaction in Table 5, which was the case (b = .03, t = 2.87, p < .01). Hence, the model’s results support this prediction. Specifically, when the campaign was for a utilitarian product, the mean purchase intent in the exposed condition was 3.09 versus 3.00 in the control condition. This difference was significant (∆ = .09, F(1, 50) = 13.89, p < .001). The treatment effect when the campaign was for a hedonic product was not significant however (Mexposed = 2.51, Mcontrol = 2.55; ∆ = -.04, F(1, 50) = 1.00, p = .32). Hypothesis 2 predicted a larger positive treatment effect for purchase intent for high-involvement products than for low-involvement products. This implies a significant positive involvement exposed

interaction in Table 5, which was found (b = .03, t = 3.13, p < .01). Accordingly, this hypothesis is also supported. Specifically, when the campaign was for a high-involvement product, the mean purchase intent in the exposed condition was 2.59 versus 2.50 in the control condition. This difference was significant (∆ = .09, F(1, 50) = 7.07, p = .01). On the other hand, the treatment effect when the campaign was for a low-

6

We also considered campaign year as a moderator of the effects of interest (e.g., treatment effects may have varied across years). We tested this with year interactions but found no evidence of significant variation in effects by year.

14

involvement product was marginally significant and negative (Mexposed = 3.01, Mcontrol = 3.05; ∆ = -.04, F(1, 50) = 2.77, p = .10). Hypothesis 3 proposed an interaction between product type and product involvement with respect to the exposed-versus-control treatment effect. Support for this hypothesis was also found: the utilitarian involvement exposed interaction in Table 5 was significant (b = .03, t = 2.88, p < .01). To further

examine this interaction we examined the estimated means and treatment effects. The only combination of type and involvement that resulted in a significant treatment effect was in campaigns for highinvolvement utilitarian products (Mexposed = 2.98, Mcontrol = 2.76; ∆ = .22, F(1, 50) = 40.80, p < .001). The treatment effects in the other three cells were not significant (all p > .20). Hypothesis 4 considered the effect of having been recently exposed to advertisements for the focal product, predicting a larger treatment effect on purchase intent when participants had not experienced (or did not recall) recent exposure. This hypothesis was also supported. The recent exposure exposed interaction in Table 5 was significant and negative (b = -.04, t = -4.15, p < .001). Specifically, when the participants recalled recent exposure the mean purchase intent in the exposed condition was 3.05 versus 3.11 in the control condition. This difference was only marginally significant (∆ = -.06, F(1, 50) = 3.05, p = .09). However, when participants did not recall recent exposure the mean purchase intent in the exposed condition was 2.55 versus 2.44 in the control condition, and the difference was highly significant (∆ = .11, F(1, 50) = 24.03, p < .001).

6.2 Robustness Checks All four of our hypotheses are supported. Importantly, our results suggest that product type and product involvement are important factors that affect whether mobile advertisement exposure increases an individual’s purchase intent for an advertised product. Consistent with our theory, the strongest positive impact of mobile display advertising appears to be found when high-involvement, utilitarian products are advertised. To check the robustness of these findings we estimated three additional random effects models, all of which are reported in Table 7. We based these models on different functional forms that correspond to recoding/transformation of the purchase intent dependent variable. First, we transformed purchase intent into a binary variable with 1 being equal to purchase intent = 5 (―very likely‖) and 0 otherwise, and estimated a random effects binary logit model (―Binary Logit 1‖). Second, we transformed purchase intent into a binary variable with 1 being equal to purchase intent ≥ 4 (―likely‖ or ―very likely‖) and again estimated a random effects binary logit model (―Binary Logit 2‖). Third, we treated purchase intent as a discrete, ordinal variable instead of continuous and estimated a random effects ordered logit model (―Ordered Logit‖).

15

Generally, the results derived from these models are largely consistent with those reported above. Hypotheses 2 and 4 are again supported under all three different model specifications. For hypothesis 1, we find support in Binary Logit 2 and Ordered Logit models, but not in Binary Logit 1 (although the sign is correct, the parameter does not reach significance). For hypothesis 3, we find support in Binary Logit 1 and Ordered Logit models, but not in Binary Logit 2 (once again, the sign is consistent but not significant). Importantly, we find complete support for all hypotheses in the ordered logit model, which is the least restrictive specification since purchase intent was used as measured. In both binary logit models purchase intent was dichotomized according to an arbitrary threshold, which is more restrictive. Thus, we are confident that our results are robust to variations in model specification.

7. CONCLUSION This paper examined the ability of 54 ―low-fidelity‖ mobile display advertising campaigns to affect consumers’ levels of purchase intent for advertised brands and products using a field experiment involving 27,753 participants who were randomly assigned to either an exposed or a control condition. By accounting for two theoretically important campaign characteristics—whether the product was hedonic or utilitarian, and whether the product was high or low involvement—and whether participants recalled recent exposure to advertising for the focal advertised product we were able to test a set of hypotheses based on the psychology literature on persuasion and information processing that predicted conditions under which low-fidelity mobile display advertisements would be effective. As predicted, for highinvolvement, utilitarian products low-fidelity digital advertisements are effective in lifting consumers’ intentions to purchase products. Overall, the results indicate low-fidelity mobile advertisements that, as a consequence of their design and technology restrictions, contain relatively small amounts of information are effective when used in situations where consumers have both the ability and the motivation to process and elaborate on the information in a deliberate fashion. Equally importantly, in situations where these underlying conditions are not met (e.g., when consumers are more inclined to rely on ―gut instinct‖ impressions in the case of hedonic products), low-fidelity mobile advertisements appear to be ineffective (at least in lifting purchase intent) because consumers are less likely to have motivation and ability to actively process the limited information contained in these advertisements. The current research is not without limitations. First, our data covered low-fidelity mobile display advertising campaigns and not other types of mobile advertising. A very promising avenue for future research on mobile advertising would entail a comparison between the effectiveness of campaigns employing exclusively low-fidelity advertising units (such as those examined here), exclusively high-

16

fidelity advertising units, and combinations of these types. However, despite the potential appeal of highfidelity mobile advertising to both advertisers and consumers (see Table 1 for examples), deployment of these units requires consumers to be using smartphones and other sophisticated mobile devices that support rich-media content (and high-speed data connections). Although access to smartphones and mobile broadband Internet is increasing, the penetration of these technologies remains very low, especially in emerging markets around the world (The Economist 2011). Hence, low-fidelity mobile advertising units, at least for the immediate future, are substantially more attractive to advertisers seeking to reach broader consumer populations. A second limitation is the use of purchase intent as a single measure of advertising effectiveness. Advertisers are often concerned with achieving other objectives, and mobile advertising campaigns may not always be designed to directly impact purchase intent and, accordingly, sales. For instance, a campaign objective may be to generate word-of-mouth (e.g., viral marketing). Notwithstanding, increasing purchase intent (and, by expected positive correlation, actual purchase behavior) is the most important campaign objective and is also related to consumers being more likely to include a given brand or product in a consideration set. A future research direction would be to examine other campaign objectives and, where possible, to link mobile advertising directly to purchase behavior. Digital advertising in both mobile and non-mobile channels is currently experiencing explosive growth. As advertisers shift larger proportions of their budgets into digital advertising channels, a more detailed understanding of the conditions under which different types of digital advertising will be effective is increasingly important. Despite a considerable amount of literature on various forms of nonmobile digital advertising, mobile advertising research has received much less attention. The current research with its thorough foundations in consumer psychology combined with field experiment data represents an important step towards a better understanding of the determinants of mobile advertising effectiveness for researchers and practitioners alike. We believe that our typology for distinguishing between low- and high-fidelity types of digital advertising units offers a new approach for classifying digital advertising. Although this paper’s focus is solely on low-fidelity mobile advertising, we hope this research encourages future work on this and related areas.

17

8. REFERENCES

Apsler, R., & Sears, D. O. (1968). Warning, personal involvement, and attitude change. Journal of Personality and Social Psychology, 9(2), 162-166. Avnet, T., Pham, M. T., & Stephen, A. T. (2012). Consumers’ trust in feelings as information. Journal of Consumer Research, forthcoming. Barwise, P., & Strong, C. (2002). Permission-based mobile advertising. Journal of Interactive Marketing, 16(1), 14-24. Chaiken, S. (1980). Heuristic versus systematic information processing and the use of source versus message cues in persuasion. Journal of Personality and Social Psychology, 39(5), 752-766. Danaher, P. J., Lee, J., & Kerbache, L. (2009). Optimal internet media selection. Marketing Science, 29(2), 336-347. Dellarocas, C., Zhang, X. M., & Awad, N. F. (2007). Exploring the value of online product reviews in forecasting sales: The case of motion pictures. Journal of Interactive Marketing, 21(4), 23-45. Dhar, R., & Wertenbroch, K. (2000). Consumer choice between hedonic and utilitarian goods. Journal of Marketing Research, 37(1), 60-71. Drossos, D., Giaglis, G. M., Lekakos, G., Kokkinaki, F., & Stavraki, M. G. (2007). Determinants of effective SMS advertising: An experimental study. Journal of Interactive Advertising, 7(2), 16-27. eMarketer. (2011). Mobile Advertising and Marketing : The eMarketer View. Epstein, S., & Pacini, R. (1999). Some basic issues regarding dual-process theories from the per- spective of cognitive-experiential self-theory. In S Chaiken & Y. Trope (Eds.), Dual process theories in social psychology (pp. 462-482). Garg, R., & Telang, R. (2011). Estimating App Demand from Publicly Available Data. Working Paper. Gartner. (2011). Forecast: Mobile Advertising, Worldwide, 2008-2015.

18

Geuens, M., Pham, M. T., & Pelsmacker, P. D. (2010). Product involvement vs. product motives as moderators of the effects of ad-evoked feelings: an analysis of consumer responses to 1,100 TV commercials. Advances in Consumer Research (Vol. 38). Ghose, A., & Han, S. P. (2011). An Empirical Analysis of User Content Generation and Usage Behavior on the Mobile Internet. Management Science, 57(9), 1671-1691. Ghose, A., & Yang, S. (2009). An empirical analysis of search engine advertising: sponsored search in electronic markets. Management Science, 55(10), 1605-1622. Goldfarb, A., & Tucker, C. (2011). Online Display Advertising: Targeting and Obtrusiveness. Marketing Science, 30(3), 389-404. Holbrook, M. B., & Hirschman, E. C. (1982). The Experiential Aspects of Consumption: Consumer Fantasies, Feelings, and Fun. Journal of Consumer Research, 9(2), 132-140. MacKenzie, S. B., & Spreng, R. A. (1992). How does motivation moderate the impact of central and peripheral processing on brand attitudes and intentions? Journal of Consumer Research, 18(4), 519529. Pappachen, G., & Manatt, K. (2008). The Mobile Brand Experience: Measuring Advertising Effectiveness on the Mobile Web. ESOMAR. Petty, R. E., & Cacioppo, J. T. (1979). Issue Involvement Can Increase Or Decrease Persuasion By Enhancing Message-Relevant Cognitive Responses. Journal of Personality and Social Psychology, 37(10), 1915-1926. Petty, R. E., & Cacioppo, J. T. (1981). Attitudes and Persuasion: Classic and Contemporary Approaches. Dubuque, IA: William C. Brown. Petty, R. E., & Cacioppo, J. T. (1983). Central and Peripheral Routes to Persuasion: Application to Advertising. In Advertising and Consumer Psychology, eds. Percy, L., & Woodside, A. Lexington, MA: Lexington Books. Petty, R. E., & Cacioppo, J. T. (1986). Communication and persuasion: Central and peripheral routes to attitude change. New York: Springer-Verlag.

19

Petty, R. E., Cacioppo, J. T., & Schumann, D. (1983). Central and peripheral routes to advertising effectiveness: The moderating role of involvement. Journal of Consumer Research, 10(2), 135-146. Petty, R. E., Ostrom, T. M., & Brock, T. C. (1981). Cognitive responses in persuasion. Hillsdale: Erlbaum. Pham, M. T., Lee, L., & Stephen, A. T. (2012). Feeling the Future: The Emotional Oracle Effect. Journal of Consumer Research, forthcoming. Rudd, J., Stern, K., & Isensee, S. (1996). Low vs. high-fidelity prototyping debate. interactions, 3(1), 7685. Schwarz, N., Bless, H., & Bohner, G. (1991). Mood and persuasion : affective states influence the processing of persuasive communications. Advances in Experimental Social Psychology, 24(1991), 161-199. Sherif, M., & Hovland, C. I. (1961). Social judgment: Assimilation and contrast effects in communication and attitude change. Oxford, England. Shim, J. P., Park, S., & Shim, J. M. (2008). Mobile TV phone: current usage, issues, and strategic implications. Industrial Management Data Systems, 108(9), 1269-1282. Stephen, A., & Galak, J. (2012). The Effects of Traditional and Social Earned Media on Sales: An Application to a Microlending Marketplace. Working Paper, University of Pittsburgh. Toubia, O., & Stephen, A. (2012). Intrinsic Versus Image-Related Motivations in Social Media: Why Do People Contribute Content to Twitter? Working Paper, Columbia University. Trusov, M., Bucklin, R., & Pauwels, K. (2009). Effects of word-of-mouth versus traditional marketing: Findings from an internet social networking site. Journal of Marketing, 73(5), 90–102. Tsang, M. M., Ho, S., & Liang, T. (2004). Consumer Attitudes Toward Mobile Advertising : An Empirical Study. International Journal of Electronic Commerce, 8(3), 65-78. The Economist (2011). Beyond the PC. The Economist, 8th October.

20

FIGURE 1 EXAMPLES OF MOBILE DISPLAY ADVERTISEMENTS

A. Nike advertisement in mobile WAP browser

B. Match.com advertisement in The Weather Channel app for iPhone

21

FIGURE 2 EXPOSED VERSUS CONTROL CONDITIONS AND SURVEY INVITATION

Exposed Condition

Survey link
Advertisement

Control Condition

Survey link

22

TABLE 1: TYPOLOGY OF DIGITAL ADVERTISING MEDIA Low-Fidelity          Can convey relatively small amounts of information to consumers Largely text based Low image resolution Static Limited interactivity Small size or size-constrained Simple Lower production cost       Examples: Non-mobile Display banners, Facebook Ads, Google Adwords, Sponsored Tweets, YouTube overlay High-Fidelity Can convey relatively large amounts of information to consumers Less reliant on text Visually appealing, including higher image resolution and/or video content Dynamic Interactive Large size and size-adjustable, including ability to dynamically expand/contract with user input Higher production cost

Defining Attributes

Interactive Flash/HTML5, full- or partscreen interstitial image or video, Pre-roll video, YouTube video Apple iAd, click-to-call/locate, games, standalone apps, video Apple iAd, clickto-call/locate, games, standalone apps, video

Mobile

Display banners, MMS, SMS

TABLE 2: CAMPAIGN SUMMARY STATISTICS

Number of campaigns Mean number of respondents per campaign (st. dev) Mean campaign length in days (st. dev) Percent of campaigns for business-to-consumer brands (versus business-tobusiness brands) Percent of campaigns for new products (versus existing products or extensions) Percent of campaigns for products (versus services) Percent of campaigns by industry: Alcohol Automotive Consumer Packaged Goods Entertainment Finance Government & Non-Profit Health & Pharmaceutical Restaurant Retail Technology & Communications 23

54 513.94 (554.87) 54.80 (32.79) 96.30% 7.41% 85.19% 3.70% 12.96% 33.33% 9.26% 16.67% 3.70% 5.56% 3.70% 1.85% 9.26%

TABLE 3: NUMBERS OF PARTICIPANTS AND CAMPAIGNS BY CAMPAIGN TYPE AND INVOLVEMENT Type Hedonic Hedonic Involvement High Low Campaigns 8 19 Participants 2,244 7,371 Represented Industries Automotive, Technology & Communications Alcohol, Consumer Packaged Goods, Entertainment, Restaurants Automotive, Finance, Government & Non-Profit, Health & Pharmaceuticals, Technology & Communications Consumer Packaged Goods, Government & Non-Profit, Retail

Utilitarian Utilitarian Totals

High Low

17 10 54

9,895 8,243 27,753

TABLE 4: NUMBERS OF EXPOSED AND CONTROL PARTICIPANTS BY CAMPAIGN TYPE AND INVOLVEMENT

Type Hedonic Hedonic Utilitarian Utilitarian Total

Involvement High Low High Low

Exposed Control Total 1,177 1,067 2,244 3,811 3,560 7,371 5,891 4,004 9,895 3,578 4,665 8,243 14,457 13,296 27,753

24

TABLE 5: RANDOM EFFECTS LINEAR MODEL FOR PURCHASE INTENT (1-5 SCALE) Full Model Estimate t-value Effects Corresponding to Hypotheses: Utilitarian × Exposed (H1) Involvement × Exposed (H2) Utilitarian × Involvement × Exposed (H3) Recent exposure × Exposed (H4) Other Effects: Intercept Utilitarian Involvement Recent exposure Exposed Utilitarian × Involvement Utilitarian × Recent exposure Involvement × Recent exposure Utilitarian × Involvement × Recent exposure Utilitarian × Recent exposure × Exposed Involvement × Recent exposure × Exposed Utilitarian × Involvement × Recent exposure × Exposed Industry fixed effects Year fixed effects New product vs. existing product fixed effects Product vs. service fixed effect B2C fixed effect Campaign length Campaign random effect -2 log likelihood AIC BIC R2
Notes: * p < .05, ** p < .01, *** p < .001

Model Without Controls Estimate t-value .03** .03** .03** -.04*** 3.45*** .24*** -.21*** .29*** .01 .03 -.01 -.01 .02 .00 .00 -.01 2.90 3.19 2.89 -4.17 57.17 4.02 -3.55 27.33 1.17 .57 -.79 -.36 1.54 .01 .28 -.40 No No No No No No 5.05 91,721.30 91,757.30 91,793.10 .45

.03** .03** .03** -.04*** 2.68*** .26** -.24 .29*** .02 .07 -.01 -.01 .02 .00 .01 -.01

2.87 3.13 2.88 -4.15 4.53 3.27 -.86 27.33 1.19 .91 -.80 -.36 1.53 .02 .27 -.38 Yes Yes Yes Yes** Yes Yes 5.06 91,698.04 91,768.04 91,837.65 .45

.10***

.16***

25

TABLE 6: ESTIMATED MEANS AND TREATMENT EFFECTS FOR PURCHASE INTENT (1-5 SCALE)

High Involvement Low Involvement

Exposed 2.98 3.19

Utilitarian Control Δ *** 2.76 .12 3.23 -.04

% 7.97 -1.24

Exposed 2.21 2.82

Hedonic Control Δ 2.24 -.03 2.86 -.04

% -1.34 -1.40

Notes: Δ = Treatment Effect = Exposed – Control. % = 100*(Exposed – Control)/Control. * p < .05, ** p < .01, *** p < .001 based on an F-test of Δ = 0

26

TABLE 7: RANDOM EFFECTS BINARY LOGIT AND ORDERED LOGIT MODELS FOR PURCHASE INTENT Binary Logit 1a Estimate t-value Effects Corresponding to Hypotheses: Utilitarian × Exposed (H1) Involvement × Exposed (H2) Utilitarian × Involvement × Exposed (H3) Recent exposure × Exposed (H4) Other Effects: Intercept Intercept 1 Intercept 2 Intercept 3 Intercept 4 Utilitarian Involvement Recent exposure Exposed Utilitarian × Involvement Utilitarian × Recent exposure Involvement × Recent exposure Utilitarian × Involvement × Recent exposure Utilitarian × Recent exposure × Exposed Involvement × Recent exposure × Exposed Utilitarian × Involvement × Recent exposure × Exposed Industry fixed effects Year fixed effects New product vs. existing product fixed effects Product vs. service fixed effect B2C fixed effect Campaign length Campaign random effect -2 log likelihood AIC BIC .03 .06** .04* -.05* -2.22* 1.35 2.96 1.94 -2.48 -2.24 Binary Logit 2b Estimate t-value .04* .04* .02 -.03* -.75 2.24 2.54 1.27 -2.01 -.87 .95* .39 .70 -1.67*** .33*** -.26** .39*** .02 .10 .00 -.02 .03 .00 .00 -.01 -2.26 -.93 1.67 3.98 -3.76 3.22 -25.11 -.97 -1.16 .04 1.04 -1.77 .07 .01 .76 Yes Yes Yes Yes* Yes Yes 5.02 302,489.40 302,555.40 302,827.03 Ordered Logitc Estimate t-value .04* .04** .04** -.05*** -2.53 -2.79 -2.35 3.51

.41** -.23 .43*** .02 .14 -.01 .01 .01 -.03 -.03 -.03

.28***

3.11 -.50 22.34 1.04 1.14 -.37 .60 .36 -1.63 -1.48 -1.68 Yes Yes Yes Yes* Yes Yes 4.92 121,584.00 121,666.00 122,003.48

.39*** -.58 .43*** .03 .14 .00 .03 .02 .01 .02 .00

3.52 -1.44 25.00 1.79 1.27 .08 1.46 1.26 .64 1.28 .04 Yes Yes Yes Yes** Yes Yes

.21***

4.95 117,016.30 117,098.30 117,435.78

.23***

27

Notes: a Binary logit 1 uses a response variable where 1 = (purchase intent = 5) and 0 = (1 ≤ purchase intent < 5). b Binary logit 2 uses a response variable where 1 = (purchase intent ≥ 4) and 0 = (1 ≤ purchase intent < 4). c Ordered logit uses purchase intent (1-5) as the response variable. * p < .05, ** p < .01, *** p < .001.

28