Seeding Viral Content The Role of Message and Network Factors

Yuping Liu-Thompkins Online viral campaigns require a seeding strategy that involves choosing the first- Old Dominion University generation consumers to spread a viral message to. Building on social-capital theory and [email protected] social-network analysis, this research examine key aspects of the seeding strategy by tracking the diffusion of 101 new videos published on YouTube. The results show that the need for a “big-seed” strategy (i.e., using many seed consumers) depends on message quality. Furthermore, one should choose consumers who have strong ties with the advertiser and who also have strong influence on others, rather than simply wider reach. Among seed consumers, they should share a moderate amount of interest overlap instead of being too homogeneous or heterogeneous as a group.

INTRODUCTION and Lilien, 2008). This view comes from the reali- “Viral marketing” refers to the act of propagating zation that viral marketing outcomes are affected marketing messages through the help and cooper- by many factors that firms have limited control ation from individual consumers. It departs from over. Although these factors introduce a great deal traditional advertising in its reliance on consumer of uncertainty into the viral-marketing process, it word of mouth (WOM) instead of mass media as does not mean that firms cannot make informed the message conveyance vehicle. decisions to maximize the possibility of success Compared with traditional advertising, viral (Kalyanam et al., 2007). marketing enjoys the benefits of lower cost, higher One such area that marketers may control is how credibility, faster diffusion, and better targeting of to start the viral diffusion process—what usually consumers (Bampo et al., 2008; Dobele, Toleman, is referred to as the “seeding strategy.” A seeding and Beverland, 2005). Furthermore, the emergence strategy involves determining how many initial of online communities and social media in recent consumers (“seeds”) are needed to disseminate years have vastly extended individual consumers’ a viral message to and what types of consumers influence beyond their immediate circle of close to choose as seeds. As these seed consumers are friends to more casual acquaintances and some- responsible for the initial dissemination of the viral times even strangers (Duan, Gu, and Whinston, message to other fellow consumers, selecting the 2008). This significantly increased the scale of viral right targets as seeds can have a significant impact marketing, putting it into a more central position on later rounds of the viral diffusion process in company strategy (Ferguson, 2008). (Bampo et al., 2008; Watts and Peretti, 2007). Despite its increasing use, both marketing prac- Academic research on online viral market- titioners and researchers have pointed out the elu- ing offers limited guidance on choosing a proper siveness of viral marketing success and a general seeding strategy. Although existing studies have lack of understanding of what drives the success of examined the impact of individual and content viral marketing efforts (Ferguson, 2008; Kalyanam, characteristics that affect the pass-along of viral McIntyre, and Masonis, 2007). Some see viral mar- information, few have explicitly addressed the keting as more of an art than a science (De Bruyn proper choice of seed consumers.

DOI: 10.2501/JAR-52-4-000-000 December 2012 JOURNAL OF ADVERTISING RESEARCH 59 Seeding Viral Content

Addressing this gap in the literature, Message characteristics relate to the consumers. The central thesis from this the current research draws from the social- content and creative design of a viral stream of research is that the structure of capital theory and social-network analysis message, which are under the control of the social network through which a viral to identify four key elements of the seed- the advertiser (Ho and Dempsey, 2010; message spreads can affect the even- ing strategy: Kalyanam et al., 2007). An effective viral tual reach and influence of the message message should break through clutter and (Bampo et al., 2008; De Bruyn and Lilien, • the number of seeds to use, consumer indifference to encourage fur- 2008). Furthermore, a consumer’s role in • the strength of tie between seed individ- ther pass-along of the message. Research- diffusion depends on his or her position uals and the message originator, ers have found, for instance, that humor in the social network as defined by the • the level of influence of individual and sex appeal are popular tactics used in consumer’s relationship with others in the seeds, and viral messages (Golan and Zaidner, 2008) network, such as network centrality and • the interest homogeneity among seed and that the social visibility of a viral mes- tie strength (Goldenberg, Han Lehmann, individuals. sage encourages its diffusion (Susarla and Hong, 2009; Kiss and Bichler, 2008; et al., 2012; Salganik, Dodds, and Watts, Susarla et al., 2012). Using viral videos from YouTube as the 2006). Research in this area has often produced backdrop, it empirically tests the relation- Besides message characteristics, indi- conflicting results. For instance, on the ship between these seeding decisions and vidual consumers also play a critical role effect of network structure, some research- viral-diffusion outcome. Because informa- in the viral marketing process. This cat- ers have shown that a scale-free network, tion for making these decisions easily can egory of influence has received the most where only a few members have many be obtained from observable online social- extensive examination in the literature. connections, facilitates social contagion network activities and internal customer Findings in this area show that consum- (Barabási, 2002; Smith, Coyle,, Lightfoot, data, the findings from this research can ers’ personality traits (e.g., Chiu, Hsieh, and Scott, 2007). Others have found no offer very practical guidance to optimally Kao, and Lee, 2007; Sun, Youn, Wu, and difference, however, between a scale-free seeding a viral-marketing campaign. Kuntaraporn, 2006); demographics (e.g., network and a random network, where Trusov, Bodapati and Bucklin, 2010); usage most network members have a similar What Affects Online Viral characteristics (e.g., Niederhoffer, Mooth, number of network connections (Kiss and Marketing Success? Wiesenfeld, and Gordon, 2007; Sun et al., Bichler, 2008). Yet, a third study concluded In the last 5 to 10 years, interest in online 2006); and motivation for sharing content that cascades were less likely to happen in viral marketing has increased among mar- (e.g., Eccleston and Griseri, 2008; Phelps a network where individual influence is keting and advertising scholars. Studies in et al., 2004) all can affect the success of highly unbalanced than in a random net- this area typically have focused on either viral messages. work (Watts and Dodds, 2007). intermediate actions/processes such as For example, researchers have found probability of opening and passing along female and younger consumers tend to Gaps in the Literature viral information (e.g., Ho and Dempsey, exert more influence on their targets and Although academic researchers have 2010; Phelps et al., 2004), or end outcomes to be more susceptible to viral influences started to construct a roadmap of fac- such as the eventual reach of a viral cam- than male and older consumers (Katona tors contributing to the success of online paign and the adoption of the promoted et al., 2011; Trusov et al., 2010). Studies viral marketing, research in this area still product (e.g., Bampo et al., 2008; Katona, also have associated both extroversion and has been very limited (Chiu et al., 2007; Zubcsek, and Sarvary, 2011). innovativeness with a higher tendency to De Bruyn and Lilien, 2008) and has pro- In answering the question of what pass along content (Chiu et al., 2007; Sun duced fragmented and sometimes con- affects viral marketing success, three types et al., 2006). From a motivational stand- flicting results. of factors have been suggested: point, research has consistently found More specifically, there are two import- altruism to drive message sharing (e.g., ant gaps in the literature that need to be • message characteristics, Ho and Dempsey, 2010; Phelps et al., 2004). addressed: • individual sender or receiver character- Though individual characteristics focus istics, and on a single consumer, network charac- • Most existing studies have relied on • social network characteristics. teristics describe the connection between computer simulations or consumer

60 JOURNAL OF ADVERTISING RESEARCH December 2012 Seeding Viral Content

surveys. Although simulation allows Specifically, researchers have called then comprise the social capital that it can controlled experimentation with net- for more analysis of a viral campaign’s draw upon for fulfilling its goals such as work properties that are difficult to seeding strategy (Bampo et al., 2008; spreading a viral message or increasing implement in a field setting, results Yang, Yao, Ma, and Chen, 2010), which brand awareness. from such studies are constrained by defines the choice of consumers that An important benefit of social capital is parameter and model assumptions that companies should initially spread the the facilitation of information flow (Lin, often prove unrealistic in the real-world viral message to. 2001; Van den Bulte and Wuyts, 2007), (Bampo et al., 2008). As a result, the As these seeds will initiate viral prop- which, in essence, is what an advertiser conclusion that a viral campaign can be agation among fellow consumers, they aims for when it launches a viral cam- more successful under a certain simu- can play a critical role in the eventual paign. To successfully propagate a viral lated condition may not mean it will success of a viral campaign (Bampo message to a wide network of consumers, happen in reality. et al., 2008; Watts and Peretti, 2007). an advertiser needs to purposefully con- Studies based on consumer surveys More research is needed to help identify struct and mobilize its social capital for partly make up for this by offering a ideal seed targets for viral-marketing optimal outcomes (Portes, 1998). closer view of consumers’ attitudes campaigns. This process involves careful selection and intentions. They suffer, however, of the initial target consumers (i.e., seeds) from noise and bias often present in Research Hypotheses to maximize access to and mobilization self-reported and retrospective data Overview of resources within the advertiser’s social (De Bruyn and Lilien, 2008). Further- Addressing the gaps in the literature, the network. In this respect, the social capi- more, previous survey studies often current research focused explicitly on the tal theory suggests three dimensions that used a rather homogeneous sample optimal seeding of viral-marketing cam- should be considered (Lin, 2001): such as college students to draw their paigns. The author drew upon the social- conclusions and they were dispropor- capital theory and its paralleling social • the extensity of ties, which is captured tionately focused on successful com- resources theory (Lin, 1999; Portes, 1998) by the size of the network (Bourdieu, munications (De Bruyn and Lilien, to identify important factors to consider 1986); 2008). This limited the generalizability when designing a seeding strategy. • relationship strength between the focal of the findings from these studies. Rec- These theories state that one can derive actor and its network connections ognizing such limitations, other stud- significant tangible and intangible bene- (e.g., Granovetter, 1973; Lin, Ensel and ies have issued a call for more research fits from one’s social network and from Vaughn, 1981); based on actual behavior of heteroge- the resources embedded in the network. • the resources embedded within the net- neous consumers in a natural setting In other words, one’s social connections work as held by its members (Lin, 1999). (Bampo et al., 2008). are a form of capital that can be utilized to The last dimension can be further bro- attain one’s goals (Coleman, 1990). ken down into the level and the diver- • Existing research often has failed to Although the relevance of social capital sity of resources possessed by network recognize the strategic nature of online to brands is less obvious and less perva- entities (Lin, 2001; Van den Bulte and viral marketing. Although online viral sive in the traditional advertising environ- Wuyts, 2007). marketing often has been viewed more ment, the one-to-one interaction between as an art than a science (De Bruyn and businesses and consumers through social Based on these dimensions, the current Lilien­, 2008), a study of an online service media has elevated the social plain for research identified four critical aspects of provider showed that companies can brands and has transformed brands a seeding strategy: tweak the inputs into a viral campaign into active participants in online social to increase the chance of success (Kaly- networks. • seed network size, anam et al., 2007). From this view, a brand (or its company) • tie strength, To aid companies in such efforts, can be considered an actor embedded in an • seed influence, which signals resource more research is needed to examine the extended network of consumers and other level, and decisions that advertisers can make in entities in the marketplace. Its relationship • seed homogeneity, which serves as an designing a viral-marketing campaign. with these consumers and other entities indicator of seed resource diversity.

December 2012 JOURNAL OF ADVERTISING RESEARCH 61 Seeding Viral Content

More specifically, when designing the Consequently, there is a tradeoff between focuses on one factor: quality of the viral seeding strategy of a viral campaign, an using many seed individuals and main- message. advertiser needs to answer four questions: taining a low viral campaign cost. In this case, ”quality” broadly is defined To determine the right balance, some as consumers’ general evaluation of the • How many seeds should be used? insight can be gleaned from epidemiology message. A high-quality message may be • Should these seed consumers have research. A key concept in that literature one that is creative, entertaining, informa- strong or weak ties with the firm/ is the basic reproductive ratio, defined as tive, or socially valuable; a low-quality brand? the expected number of secondary infec- message, conversely, may be one that fails • Is it superior to use seed consumers who tions an infected individual will cause to pique interest among consumers. In the are social hubs with a large number of (Heffernan, Smith, and Wahl, 2005). When former situation, the pass-along rate (hence connections with other consumers? this ratio is smaller than one, the network basic reproductive ratio) is likely to be high, • Should seed consumers be chosen will show subcritical growth (Bampo and the number of seed consumers becomes from a heterogeneous or homogeneous et al., 2008), and the disease will wane out less important; in the latter situation, the population? without saturating the population. When number of seed consumers may determine the ratio is greater than one, by contrast, the final outcome of a viral campaign. By addressing these questions and empiri- a truly viral process is established, and This leads to the first two hypotheses: cally testing the effects of these decisions exponential growth will be experienced on actual diffusion outcomes, the current through generations of the disease propa- H1: The number of seeds will have a research aims to provide a systematic gation process. positive effect on the diffusion of guide to choosing the best seed consum- The larger the ratio, the more likely an a viral message. ers for initializing a viral-marketing epidemic will occur and affect the entire campaign. population (Heffernan et al., 2005). The H2: The relationship in H1 will be relevance of this basic reproductive ratio stronger when the viral message Number of Seeds to the seeding decision lies in its impact on quality is low than when the When picking the seed individuals to how important the size of the initial seed message quality is high. spread a viral message, a natural first con- group is. When the ratio exceeds one and sideration is how many seeds to use. exponential growth ensues, having many Strength of Tie with Seeds From a social-capital standpoint, initial seeds is not critical. Besides choosing the right number of the more network connections that are The main consideration in this context seeds to start a viral campaign, it is also mobilized in a given situation, the more is to have enough seeds to ensure that the important to consider the strength of con- resources will be available for the focal propagation does not stop early (Bampo nection a viral content creator has with actor to utilize in achieving its objectives et al., 2008). When the ratio is smaller than seed consumers. For instance, in the case (Burt, 1997; Lin, 1999). This supports the one, however, the size of the seed group of viral brand messages, companies may use of a large number of seeds. From a becomes much more important and can want to consider tie strength as measured mathematical perspective—other things determine the final reach of a disease by brand loyalty or brand usage. Accord- being equal—the larger the number of or campaign. In such situations, there ing to the social-capital theory, strong ties seeds, the more opportunity there is for is research that advocates a “big-seed” provide ecological reasons for network a message to reach other consumers and approach, where a large number of seeds members to lend resources to others, not the more likely the message will create an are used (Watts and Peretti, 2007). necessarily for direct gain from the bor- impact. This is the basic idea behind mass- Applying the preceding to a viral mar- rower but for reputation and other ben- media advertising and explains the popu- keting message, the importance of the efits that one can derive from the entire larity of advertising during major events initial seed group size may be contin- network (Burt, 1997; Lin, 2001). As a such as the Super Bowl. gent on the likelihood of seed consumers result, network members who are con- The downside to using many seeds, to pass along the viral message to other nected to the focal actor through a strong however, is the high cost associated consumers. Although various message tie will be more motivated to cooperate with the strategy. This partly defeats the and individual characteristics can affect with the actor than those connected via a cost-effective nature of viral marketing. this likelihood, the current research weak tie.

62 JOURNAL OF ADVERTISING RESEARCH December 2012 Seeding Viral Content

More specifically, in the context of an It should be noted, however, that the online viral-marketing campaign, a few advantages can result from having a social‑capital literature also has suggested strong tie with seed consumers: a critical role played by weak ties. • In today’s already-cluttered online envi- ronment, information shared through a strong tie is more likely to be noticed. Although such results seem to suggest creator will lead to more suc- Either due to higher interest or due to that one should select consumers with cessful diffusion than individu- social pressure, consumers are more weak ties to the company as viral cam- als who have weak ties with the likely to open messages sent from an paign seeds, a few issues undermine the message creator. entity that they feel close to (De Bruyn strategic appropriateness of this decision. and Lilien, 2008). First, as the authors pointed out, the Seed Influence • For similar reasons, stronger ties can results do not reflect the impact of over- As seed consumers start to pass along a also increase the possibility that the all WOM but rather the incremental viral message to fellow consumers, the message will be passed along to others influence of WOM by the sample consum- extent of influence each seed has can play (Chiu et al., 2007), which is crucial to ers beyond existing WOM (Godes and an important role in further spreading of starting later stages of the viral process. Mayzlin­, 2009). As loyal customers likely the message. • When persuasion is the goal, informa- already spread words about the restau- This level of influence represents the tion shared through a strong tie tends to rant, their true impact is likely to have social capital a seed consumer possesses be more persuasive and, therefore, can been underestimated. that an advertiser can indirectly leverage have a larger influence on the recipi- Second, non-customers have not expe- in its viral campaign (Van den Bulte and ent (Bansal and Voyer, 2000; Sun et al., rienced the company that is being pro- Wuyts, 2007). A frequently used proxy for 2006). These advantages suggest the moted. As a result, their testimonials may influence is the number of connections superiority of choosing seed consumers be considered less credible and trustwor- an individual has (Wasserman and Faust, who have a strong tie with the firm. thy than those conveyed by consumers 1994), which has been shown to follow a who have a strong tie with the company. power-law distribution in online networks It should be noted, however, that the Third, as consumers with no (or weak) (Barabási, 2002). In a power-law distrib- social-capital literature also has sug- ties to a company have low motivation to uted network, a small number of nodes gested a critical role played by weak ties spread words about the business, the com- have disproportionately large numbers of (Granovetter, 1973): The argument is that pany may need to provide extra financial connections, whereas the rest have only a weak ties connect network clusters that incentive to encourage these consumers’ small number of connections. The former would otherwise be isolated from one participation in the viral process. In the forms the hubs of the network. another and, as a result, can increase the case of the foregoing study, BzzAgent Given the large disparity between hubs reach of a viral message (Godes and May- compensated the non-customer sample for and non-hubs, a key question is which zlin, 2004). A field experiment compared participation in the panel. This can make type of these consumers is better suited for the effectiveness of spreading WOM the use of weak ties a more costly strategy. seeding viral content. The answer to this through a restaurant’s loyalty-program For these reasons, the current research question is not exactly straightforward: members (i.e., regular customers) versus argues that, for the purpose of initially The well known two-step flow model of non-customers recruited from a third- seeding viral content, it still is more effect- communication emphasizes the role of party panel by BzzAgent (Godes and ive to select consumers who have stronger influential individuals in propagating Mayzlin, 2009). The authors concluded ties to the content generator than consum- information to a wider audience (Katz and that WOM initiated by non-customers ers with weak ties. Lazarsfeld, 1955). This view was echoed (i.e., weak company-consumer tie) created This leads to the third hypothesis: in previous research on the viral diffu- more incremental impact than that initi- sion of innovation (e.g., Goldenberg et al., ated by customers (i.e., strong company- H3: Seeding individuals who have 2009; Rogers, 1962). The basic argument consumer tie). strong ties with the message is that the more individuals that a seed is

December 2012 JOURNAL OF ADVERTISING RESEARCH 63 Seeding Viral Content

connected to, the more people the seed can Seed Homogeneity content consumption behavior by seed reach and potentially influence, creating Network “homogeneity” is the degree to consumers. This helps avoid potential what Van den Bulte and Joshi (2007) call a which members of a network are similar recall error or bias that may be present in “multiplier effect.” to one another. Just as birds of a feather one-sided reports of perceptual affinity. This well-established argument has flock together, researchers have found Regarding the consequence of network been challenged by recent studies, how- a tendency for humans to connect with homogeneity, past research has produced ever. Specifically, one study contended others who are similar to them, a phe- inconsistent findings: Some studies have that, instead of relying on influentials, it nomenon called homophily (McPherson, argued that similarities among individuals was more important to have a large mass Smith-Lovin and Cook, 2001). In an online may facilitate information flow, as shared of easily influenced individuals for viral community, homophily is likely to occur values and experiences among these indi- diffusion to succeed (Watts and Dodds, as well. Users with similar backgrounds viduals encourage more frequent and eas- 2007). The simulation in that research and tastes are likely to seek out and con- ier interaction with each other (McPherson showed that the cascade window—a sume similar content and, as a result, are et al., 2001; Watts, 2003). In support of this region in which large-scale diffusion was more likely to know and connect with theory, within the marketing literature, the likely to occur—resided at a rather low (or each other. homogeneity of a social system has been moderate) average number of connections. Network homogeneity can be assessed found to expedite diffusion and increase Several researchers have provided a in multiple domains. Partly due to opera- eventual market size (Gatignon, Eliash- theoretical explanation for why hubs are tional simplicity, most research has relied berg, and Robertson, 1989). not necessarily better: Due to the cost of on demographic and socioeconomic vari- More recent studies, however, have maintaining a large network, individuals ables to define similarity (e.g., De Bruyn questioned this conclusion. Data from with many connections on average have and Lilien, 2008; Kalmijn, 1998; Louch, an online travel agency’s viral campaign weaker connections, which results in less 2000). In an online community, however, demonstrated that diffusion speed was impact on others that are connected to them some of these demographic variables negatively affected by homogeneity (Lee, (Katona et al., 2011; Smith et al., 2007). either become less relevant (e.g., geo- Lee, and Lee, 2009). In a cross-cultural set- This weaker relationship can be espe- graphic distance) or are often unknown ting, income homogeneity in a country cially detrimental in spreading viral mar- (e.g., age and profession) to other indi- was shown to lead to a lower diffusion keting messages, due to the large number viduals participating in the community. rate (Van den Bulte and Stremersch, 2004). of messages circulating online. As one Instead, users tend to be guided by their In line with these findings, another study focus-group discussion revealed, individ- mutual interest in certain topics such as found that demographically similar ties uals receiving a pass-along message often sports, politics, or humor. decreased the effectiveness of viral mes- felt irritated or angry for their wasted Past research further has suggested that sages in terms of awareness, interest, and time (Phelps et al., 2004). Therefore, when such deeper-level similarities are more adoption (De Bruyn and Lilien, 2008). someone with a large number of connec- important in a group setting than surface- This latter group of findings can be tions tries to pass along a viral message level similarities characterized by factors explained by the social-capital theory, to his or her weak connections, the mes- such as race and gender (Phillips, North- where having diverse embedded resources sage will be less likely to be relevant to the craft and Neale, 2006). For this reason, the in a network is considered to increase social recipient and more likely to be ignored. current research focuses on homogeneity capital and improve the chance of finding For these reasons, the author expects a as defined by the level of shared interest the right resources needed to achieving negative relationship between the number among seed consumers. This resembles one’s goals (Burt, 2005; Lin, 1999). of connections a seed consumer has and the concept of perceptual affinity, which Reconciling the disparate findings, the outcome of viral diffusion. refers to individual similarity in personal the current research argues that the This leads to the next hypothesis: values, experiences, and tastes (De Bruyn impact of seed homogeneity on diffusion and Lilien, 2008). success does not follow a linear relation- H4: The number of connections seed However, instead of using perceived ship. Instead, it is an inverted U-shaped consumers have will have a neg- affinity as reported by survey data in that pattern, where both low and high levels ative effect on the diffusion of a study, the current study derives homoge- of homogeneity can be detrimental to viral message. neity from interests manifested in actual diffusion.

64 JOURNAL OF ADVERTISING RESEARCH December 2012 Seeding Viral Content

At very low levels of homogeneity, the Data and Model a viral marketing video is less about network consists of an eclectic and some- Study Context and Sample “creating advertising” than about “cre- what random set of connections. For these To examine empirically the impact of seed- ating something people want to watch” heterophilous individuals, group identifi- ing strategy, this research used the context (Angwin, 2009). Consequently, a suc- cation, tie strength, and stability are low of viral videos posted on YouTube. You- cessful viral marketing message is likely (Lee et al., 2009), and it is more difficult Tube is a leading online community for to resemble more closely an ordinary to effectively reinforce social norm (Alge- sharing videos and has been the origina- YouTube video than a traditional adver- sheimer, Dholakia and Herrmann, 2005; tion point for many successful viral vid- tisement. This makes knowledge about Horne, 2008). Network members do not eos in the past. YouTube videos offer a how a video on YouTube gains popu- have strong incentives to pass along any rich variety of primarily user-generated larity relevant and important to viral particular message to others within or content. Consequently, it might seem that marketers. outside the network, which impedes the studying the diffusion of YouTube videos flow of information and the diffusion of would have limited value to a commercial • The commercial relevance of the You- viral messages. Along this line, network entity (i.e., an advertiser). Tube context also is accentuated by science suggests that highly diverse net- There are, however, a number of reasons advertisers’ increasing encouragement works make it difficult for information to why the opposite may be the case: of user-generated content related to reach its destination, even when available their brands (Van den Bulte and Wuyts, network connections are present (Watts, • YouTube is an important social-media 2007). For instance, video contests fre- 2003). channel for many businesses, capturing quently are offered by brands on You- A highly homophilous group, by as many as 50 percent of Fortune Global Tube and other social-media channels, contrast, consists of individuals with 100 companies (Burson-Marsteller, which invite consumers to make their highly similar interests, which fosters 2010). Faced with the same universe of own videos on a given brand theme and group identification and makes individ- YouTube users, advertisers are subject as a return earn prizes and sometimes uals more susceptible to peer influence to the same diffusion channel and simi- fame. (Hovland, Janis, and Kelley, 1953). When lar scrutiny as other YouTube videos. Content from such programs as a message traverses across the network, What is especially relevant is the same Dunkin’ Donuts’s “How Do You Keep individuals are motivated to pass on the reliance on consumer WOM and social America Running?” YouTube contest information, either due to personal inter- contagion to spread the message. (Greenberg, 2008) often is used later est in the content topic or due to group in the sponsoring brand’s advertising norm. • Social media such as YouTube have materials—a phenomena that, again, At the macro-level beyond the ini- blurred the line between commercial demonstrates the increasingly blurred tial network, however, because of these messages and user-generated content line between commercial content and individuals’ tendency to attach to others and thereby have created a need to shift user-generated content. highly similar to them, the message is from traditional communication to a new, likely to get stuck in the circle of other more consumer-focused communication For these reasons, it is not surprising that similar individuals. This prevents the con- paradigm (Mangold and Faulds, 2009). YouTube frequently has been used as the tent from reaching a larger, more diverse In this space, advertisers have been backdrop for studying online viral and universe of users (Brown and Reingen, advised to tone down the “commercial” social-media marketing (e.g., Campbell, 1987). component of their messages (Kaplan Pitt, Parent, and Berthon, 2011; Susarla This leads to the next hypothesis: and Haenlein, 2010). Instead of a well- et al., 2012). crafted professional message, adver- When a video is posted on YouTube, H5: Seed homogeneity will have tisers may be better off acting as an the video poster’s immediate connections an inverted U-shaped effect on ordinary participant in the social con- (subscribers and friends) are the first noti- viral diffusion, with moderate versation and use subtler marketing in fied of the new addition, either through homogeneity leading to bet- their messages. e-mail or on the homepage when they ter diffusion than low and high As aptly described by a successful next visit YouTube. These immediate con- homogeneity. viral marketer, the process of making nections, in all practicality, function as

December 2012 JOURNAL OF ADVERTISING RESEARCH 65 Seeding Viral Content

seed consumers for the video. Although • as a subscriber (one-way connection to where G represents the full set of sub- one may argue that the video poster did the video poster) and scriptions by all seed consumers. not consciously choose these individuals • as a friend (two-way connection). Based on this dyadic interest homo­ as seeds, the difference in this initial audi- phily, one can then calculate the homoge- ence across videos can reveal important Because establishing a friend connection neity of a seed network as in equation (2) information about the impact of seeding requires explicit approval of the video (Wasserman and Faust, 1994). It repre- variation on video diffusion if the video poster, it is reasonable to assume that this sents the extent of subscription overlap poster had intentionally chosen these type of connection represents a stronger among video i’s seed consumers as a seed consumers. This approach especially relationship tie with the video poster than portion of all possible overlaps across is useful given the practical difficulty of that of a subscriber connection. Conse- the entire subscription set. experimenting with multiple seeding quently, the proportion of seed consum- NumSeedsi NumSeedsi strategies in a self-contained setting. ers who are connected to a video poster as ∑ ∑ Homophilyjk j=1 kj≠ In the current study, an initial pool of friends is used in the current study to indi- Homogeneityi = GG*( − 1) 105 videos was sampled over 7 days to cate the average tie strength between the (2) avoid systematic bias that may be asso- seeds and the video poster (TieStrength). ciated with a particular day of the week. Seed influence is operationalized as the A few control variables also are included Each day, a random sample of 15 videos number of individuals who are in turn con- in the analysis to account for the impact was drawn from the list of new videos nected to each seed consumer. An average of other demographic, historical, and con- uploaded to YouTube on that day using of this measure across all seed consumers tent factors. For historical influences, this the systematic sampling approach.1 of a video indicates average seed influence research controls for a video poster’s past For each video sampled, information for the video (SeedConnection). For video experience with two variables: about the video poster’s network struc- quality, YouTube allows users to rate each ture, past experience, and demographics video on a five-point scale. This research • the volume of past contribution as sig- was collected. Four videos missed com- used the average rating at the end of the naled by the number of videos posted in plete network and demographic informa- observation period as a proxy of quality the past (Vol), and tion and, therefore, were removed from for each video (Quality). • the popularity of past contributions as further analysis, resulting in an actual For network homogeneity (Homoge- measured by the average number of sample size of 101 videos. Each of these neity), this research focused on interest views across past videos (AvgView). videos was tracked on a daily basis over homophily, which—in the YouTube con- the course of 60 days. Every day, the num- text—could be gauged through shared These two variables control for the pos- ber of cumulative views and the ratings subscriptions. Consumers who share com- sibility that a video poster’s reputation for each video were recorded. mon interests with one another are likely and/or consistent success with past to have many overlapping subscriptions materials may affect the diffusion out- Variable Operationalization with one another. To derive a measure of come of the new video. The current study A video’s seed network consists of indi- interest homophily, the current research also included a video’s lagged rating viduals who are directly connected to the drew upon the idea of an affiliate network (LagRating)­ to control for the influence of video poster. The size of this network, from social-network analysis (Wasserman existing ratings on diffusion. therefore, defines the total number of and Faust, 1994): In addition, because a video can be dis- seeds (NumSeeds). On YouTube, there are covered from sources other than the seed two ways to connect to an individual: Let Subjg be a binary variable indicat- consumers (e.g., through site searches), ing individual j’s subscription status the author controlled for this influence by to channel g. We can define the level considering other sources that had led to 1 More specifically, the new videos that were added to of dyadic interest homophily between significant traffic to the video. This infor- YouTube­ on each day were ordered by their uploading time, and one video from the list was randomly picked as the start- individuals j and k as: mation is listed on YouTube as “signifi- ing point. From that starting point, every nth video was cant discovery events” and includes the sampled from the list. The exact interval (i.e., n) used for G date on which the first referral from each each day was determined by dividing the number of videos Homophilyjk = ∑Subjg Subkg (1) posted on that day by the daily sample size of 15. g=1 significant source occurred. The current

66 JOURNAL OF ADVERTISING RESEARCH December 2012 Seeding Viral Content

study counted the number of such events multiplicative form as: proposed model showed a significantly during the 60-day period (OtherSrcs) and better fit χ( 2 = 537.63, p < 0.001). dR (t) = exp(b′X (t))*dR (t) (3) included it in the model. i i 0 To check the robustness of the model,

For demographic influences, this where R0(t) is an unspecified continuous the author compared the predicted versus study considered a video poster’s age function and Xi(t) is a vector of time- actual cumulative views of the holdout (PosterAge)­ and gender (PosterFemale) and independent or time-varying covariates sample during each time period. These the corresponding seed network’s average discussed in the last section. predicted views correlated highly with the age (SeedAge) and proportion of females observed views, ranging from 0.76 to 0.97 (SeedFemale),­ thus allowing gender and This research estimated the model using for the holdout videos, indicating a good age differences in viral diffusion. Another the approach recommended by Lin et al. fit of the model. The mean correlation factor the study controlled for was the (2000), allowing arbitrary and complex coefficient was 0.91. category that a viral video was posted dependence structures among recur- under. As some topics inherently were rences, thereby permitting future views of Hypotheses Testing more appealing than others (e.g., enter- a video to depend on past events in many As the covariates were entered into the tainment content may inherently appeal to ways. model as exponentials (Table 1), the per- more people than science-related content), cent change in diffusion rate due to one a video’s category could have affected its Results unit change in a covariate was indicated diffusion potential. Ninety-one of the sample videos were by 100 × [exp(β) – 1]. used to estimate the model and the rest On the effect of seed network size (H1), The Model were used as a holdout sample. Compared as expected, the total number of seeds As the number of views for viral videos with an unconditional model that did for a video had a positive effect on the very likely contains multiple views from not have any explanatory variables, the video’s diffusion (β = 0.0017, p < 0.001). the same individual (and these views may not have a systematic pattern), the cur- Table 1 rent study used the proportional rates/ Parameter Estimates from the Model means model developed in the biomet- Covariates Unstandardized β Standard Error p rics field (Lawless and Nadeau, 1995; Lin, NumSeeds 0.0017 0.0001 <0.001 Wei, Yang, and Ying, 2000)—a frequently i used model for studying event recurrence. NumSeedsi*Qualityi –0.0006 0.0001 <0.001 Using a multiplicative formulation similar TieStrengthi 0.176 0.083 0.034 to proportional hazard models, it offers SeedConnection –0.00004 0.00001 <0.001 an efficient and parsimonious way to cap- i Homogeneity 0.518 0.168 0.002 ture the effects of covariates and provide a i 2 mechanism for inferring event recurrence Homogeneityi –1.182 0.324 <0.001 (Lin et al., 2000): Voli –0.0008 0.003 n.s.

AvgViewi 0.0003 0.00002 <0.001 Specifically, let Viewi(t) be the cumula-

tive view of video i up to time t, and LagRatingsi(t–1) 0.003 0.014 n.s. let dView (t) be the increment in views i OtherSrcsi 0.007 0.002 <0.001 over a small time interval [t, t + dt]. The PosterAgei –0.003 0.001 0.024 rate function is defined as the expec-

SeedAgei –0.021 0.003 <0.001 tation of dViewi(t) given the observ-

able history of the video and a set of PosterFemalei –0.271 0.057 <0.001 covariates that may affect recurrence. SeedFemalei –0.256 0.067 <0.001 Similar to the proportional hazard –2LL 58178.69 formulation, the proportional rates model presents the recurrence rate in a AIC 58228.69

December 2012 JOURNAL OF ADVERTISING RESEARCH 67 Seeding Viral Content

Each 100 extra seeds contributed to about that cascading is most likely with low to may have had limited influence on viral a 17.01-percent increase in diffusion rate. moderate number of connections. diffusion. This effect was qualified by a significant H5 predicted an inverted U-shaped For demographic factors, both the post- negative interaction between video qual- relationship between viral diffusion er’s age (β = –0.003, p = 0.024) and the aver- ity and total number of seeds (β = –0.0006, and seed homogeneity as measured by age age of seed individuals (β = –0.021, p < 0.001). interest homophily. This was confirmed p < 0.001) had a negative effect on diffusion To help interpret this interaction, the by a significant positive coefficient for rate. With regard to gender, videos posted procedure recommended by Aiken and the seed homogeneity variable (β = 0.518, by females experienced slower diffusion West (1991) was followed, which involved p = 0.002) and a significant negative coef- than those by males (β = –0.271, p < 0.001). calculating the simple slope for NumSeeds ficient for its quadratic term β( = –1.182, Similarly, the proportion of females among at video quality one standard deviation p < 0.001). seed individuals had a negative effect on (SD = 2.28) above and below the mean As interest homophily improved, diffu- diffusion rate (β = –0.256, p < 0.001). These (M = 522.5). sion rate first increased and then decreased findings were opposite to previous studies When video quality was low, 100 addi- after it passed a certain threshold. Using that found female consumers to be more tional seeds contributed to as much as a the estimated parameters, the optimal influential in viral diffusion (Katona et al., 30.72-percent increase in diffusion rate. interest homophily level could be calcu- 2011; Trusov et al., 2010), suggesting that When video quality was high, by con- lated as 21.91 percent. As this variable by the context of the study matters. trast, each 100 extra seeds led to only a definition lies between 0 and 1, this means Overall, the results of the current study 3.32-percent increase in diffusion rate. that about one-fifth of interest/subscription suggest more favorable influence from This was consistent with the prediction in overlap leads to optimal diffusion for viral younger consumers and males, at least H2 that the need for a large seed network videos. in the context of online viral videos. For becomes less important as video quality content category, music was used as the improves. Effect of Control Variables benchmark category. Compared with this On the effect of tie strength, it was pre- The current model controlled for three benchmark category, the entertainment, dicted that using seed individuals who types of influences: people-and-blogs, and gaming catego- have a strong tie with the content origi- ries showed significantly higher diffusion nator will facilitate viral diffusion (H3). • historical, rates, whereas the more specialized cat- This was confirmed by a significant posi- • demographic, and egories of how-to/style and pets/animals tive coefficient for TieStrength (β = 0.176, • content influence. had significantly lower view growth rates. p = 0.034). As this variable captured the The rest of the categories were not signifi- proportion of seed individuals connected For historical influence, the number of vid- cantly different from the music category. to the video poster through a stronger eos a video poster previously had posted friendship link, increasing the proportion did not have a significant effect on the dif- Discussion by 10 percent would lead to a 1.92-percent fusion of the current video (β = –0.0008, Although online viral marketing presents increase in diffusion rate. p = 0.79). The success of past videos as meas- a great opportunity for advertisers, success Concerning the impact of secondary ured by their average number of views, in this area remains elusive to most firms connections that a seed consumer has however, had a positive effect (β = 0.0003, (Ferguson, 2008; Kalyanam et al., 2007). (H4), the result confirmed a significant p < 0.001). As expected, the number of This is partly due to the many uncontrol- negative effect (β = –0.00004, p < 0.001). external sources that led to significant lable factors in the online environment. As mentioned earlier, a larger number of views of a video had a significant positive To run an effective online viral- connections represent a wider reach but effect (β = 0.007, p < 0.001). marketing campaign, it is important to potentially weaker ties between a seed The effect of lagged rating was not sig- recognize these uncertainties while at the and his or her friends. This suggests that nificant β( = 0.003, p = 0.83). This may have same time realize the ability of the firm to it is the quality rather than mere quantity been because these were newly posted make strategic choices that can maximize of influence that counts. Overall, the result videos. Especially at the beginning of the chance of success. here is consistent with the conclusion from the tracking period, rating information Taking one step in this direction, the the Watts and Dodds (2007) simulation could have been sporadic and, as a result, current research treats viral marketing as a

68 JOURNAL OF ADVERTISING RESEARCH December 2012 Seeding Viral Content

strategic process and addresses the impor- a result limited influence on subsequent social capital for the brand. There is cur- tant issue of seeding a viral-marketing generations of consumers. Therefore, rently very limited understanding of campaign, which has seldom been con- our results offer empirical support to the how advertisers can effectively build sidered in previous research. It enriches notion that, to achieve successful diffu- their social capital. This can be a fruitful the online viral marketing literature by sion, it is better to have a large number of area for future research. linking seeding choices to actual diffusion easily influenced individuals than to have outcomes. In doing so, it extends existing a few highly connected hubs in a social • The current research used existing con- studies that often relied on mathematical network (Watts and Dodds, 2007). nections and network characteristics simulation or consumer self-reported data, Finally, the current research considered on YouTube. The advantage of this which may (or may not) have reflected the desired level of seed homogeneity for approach is that it can capture actual real-world consumer reaction to possible viral marketing success. It extended previ- diffusion outcomes of a large number firm actions (Bampo et al., 2008). ous research by focusing on the interest of viral messages, hence remedying the Moreover, by including a diverse sam- homophily among seed consumers and tendency in previous studies to rely ple of both popular and unpopular videos, reconciled conflicting findings in this area purely on simulation or consumer self- the current research responds to calls to by revealing an inverted U-shaped rela- report data. By focusing on YouTube, extending existing research beyond suc- tionship between homophily and diffu- however, it neglects other processes that cessful communications in order to gen- sion outcome. may have contributed to the diffusion eralize existing findings (De Bruyn and When seed consumers share too few or of the videos. It also does not consider Lilien, 2008). too many common interests, diffusion out- other viral forms such as those distrib- The results of this study suggest that a come is not optimal. Instead, a moderately uted via Twitter or blogs. positive outcome is more likely if more heterogeneous group of consumers can Future research needs to replicate seeds are used to start a viral campaign, best increase the reach of a viral message the current findings in other settings. It the “big-seed” strategy proposed by Watts to more diverse consumer populations. would be especially useful to conduct and Peretti (2007). field experiments to compare the rela- This “big-seed” strategy, however, is not Limitations and Future Research tive effectiveness of different seeding necessary at all times. As the general qual- The current study has a few limitations that strategies. ity of the viral message improves, the need should be addressed in future research: to use a large number of seeds diminishes • The current research examined only a significantly. This points to an alternative • From a theoretical standpoint, the cur- narrow range of variables representing solution for firms that may not be able to rent research focused on the mobiliza- the four seeding decisions. This needs to afford the high cost of reaching a large tion of social capital in a viral-marketing be expanded in future research. number of seed individuals. campaign. However, the social capital- For instance, from a firm’s perspec- Regarding the type of seed consumers theory suggests that, before being able tive, a more desirable indicator of tie to use, the current results showed that it to mobilize social capital, one first needs strength may be how loyal the consumer is best to start a viral campaign with con- to deliberately construct and build social is to the firm. As another example, sumers who have a strong tie with the viral capital through economic, cultural, and instead of relying on public ratings to message originator. As these consumers social investments (Portes, 1998). judge message quality, which is avail- are likely to be more strongly influenced From this perspective, in addition able only after a message is released, it by the message originator, using these to an effective seeding strategy, a suc- may be better to gauge quality with con- consumers as seeds increases the probabil- cessful viral campaign requires long- sumer pretests. ity that the message will be passed along term investments from advertisers—in The seed influence factor also can be to further waves of consumers. addition to an effective seeding strat- extended to include not only the num- Moreover, the current analysis showed egy—to build their social capital. For ber of individuals that a seed consumer that it is not ideal to use seed consumers instance, active participation in social can potentially reach but explicit meas- with a large number of connections. The media through Twitter and Facebook ures of connection quality and how social cost of maintaining a large network can strengthen the tie between a brand much real influence a seed consumer leads to weaker average connection and as and its customers, leading to increased has on others. These extensions will

December 2012 JOURNAL OF ADVERTISING RESEARCH 69 Seeding Viral Content

enrich the current framework to offer a activities or firm internal records. For include Internet marketing, loyalty programs, and more comprehensive guide to optimally example: customer loyalty. Dr. Liu-Thompkins’s works have been seeding a viral campaign. published in the Journal of Advertising Research, • Tie strength can be captured by an indi- Journal of Marketing, Journal of Advertising, and • The current research tracked the sam- vidual’s customer status; Journal of Interactive Marketing, among others. ple videos for only 60 days. This was • within the context of online social net- based on historical evidence that many works such as Twitter, tie strength also successful viral videos gained popular- can be gauged by the frequency of inter- References ity in a very short period of time. The action the individual has with the firm;

Old Spice viral campaign in 2010, for • the reach of a consumer also can be Aiken, L. S., and S. G. West.  Multiple Regres- instance, garnered more than 5 mil- measured by the number of followers or sion: Testing and Interpreting Interactions. Thou- lion views in the first 3 days (Newman, friends he or she has; sand Oaks, CA: Sage Publications, 1991. 2010). However, in reality, it could take a • when such information is unavailable to

while for some videos to take off, a time the firm, a straightforward survey can Algesheimer, R., U. M. Dholakia, and A.

lag that may not be captured in the cur- be used to separate hubs from average Herrmann. “The Social Influence of Brand Com- rent study’s 60-day time horizon. Future individuals (Gladwell, 2000); munity: Evidence from European Car Clubs.” research should incorporate longer- • interest homogeneity can be measured Journal of Marketing 69, July (2005): 19–34. term observations to test the stability of by observing consumers’ online con-

the results here. tent consumption such as Twitter fol- Angwin, J.  “Recipe for a Successful Viral lowings, blog readings, or YouTube Video Campaign.” Wall Street Journal, 2009. Managerial Implications subscriptions. Retrieved from http://online.wsj.com/article/ Though online viral marketing often has SB123481783053894227.html. been considered either a hit-or-a-miss Because of the easy availability of infor- exercise that largely depends on luck, a mation, the seeding strategy proposed Bampo, M., M. T. Ewing, D. R. Mather, D. key conclusion from the current research here offers a practical starting point to Stewart, and M. Wallace. “The Effects of the is that firms can—and should—treat it as a designing and running viral-marketing Social Structure of Digital Networks on Viral strategic process. campaigns. Marketing Performance.” Information Systems This research examined the seeding In conclusion, even though advertisers Research 19, 3 (2008): 273–290. strategy, an important choice that involves have considerably less control when run- selecting the right consumers to initialize ning a viral-marketing campaign, there are Bansal, H. S., and P. A. Voyer.  “Word-of- the viral diffusion process. The study’s decisions that they can make to maximize Mouth Processes within a Services Purchase findings showed that the number of seeds the possibility of success. Decision Context.” Journal of Service Research 3, needed is contingent on the quality of the It is hoped that, by highlighting the stra- 2 (2000): 166–177. viral message. With a high-quality mes- tegic nature of viral marketing, the cur- sage, it is not necessary to have a sizable rent research will stimulate more scholarly Barabási, A.-L. Linked: The New Science of Net- seed network. Furthermore, it is prefer- work on how advertisers can design and works, Cambridge, MA: Perseus Publishing, able to choose seeding consumers who manage their viral-marketing campaigns 2002. have strong ties with the firm and who do more effectively. not have an extraordinarily large number Bourdieu, P. “The Forms of Capital.” In Hand- of connections. Among seed consumers, it Yuping Liu-Thompkins is associate professor of marketing book of Theory and Research for the Sociology of is best if they share a moderate amount of and E. V. Williams Faculty Fellow in the College of Education, J. G. Richardson, ed. Westport, CT: interest and are not too homogeneous or Business and Public Administration at Old Dominion Greenwood Press, 1986. divergent as a group. University in Norfolk, Virginia. She received a PhD

In applying the findings from the degree in management, with a marketing concentration Brown, J. J., and P. H. Reingen.  “Social Ties current study, it is worth noting that infor- from Rutgers University in 2002. Dr. Liu-Thompkins’s and Word-of-Mouth Referral Behavior.” Jour- mation about the factors can be easily research focuses on the intersection among marketing, nal of Consumer Research 14, December (1987): observed through online social network technology, and psychology. Her main research areas 350–362.

70 JOURNAL OF ADVERTISING RESEARCH December 2012 Seeding Viral Content

Burson-Marsteller. “The Global Social Media Eccleston, D., and L. Griseri.  “How Does Heffernan, J. M., R. J. Smith, and L. M.

Check-Up: Insights from the Burson-Marsteller Web 2.0 Stretch Traditional Influencing Pat- Wahl. “Perspectives on the Basic Reproductive Evidence-Based Communications Group,” terns?” International Journal of Market Research Ratio.” Journal of the Royal Society Interface 2, 4 2010. Retrieved from http://www.burson- 50, 5 (2008): 591–161. (2005): 281–293. marsteller.com/Innovation_and_insights/ blogs_and_podcasts/BM_Blog/Documents/ Ferguson, R.  “Word of Mouth and Viral Mar- Ho, J. Y. C., and M. Dempsey.  “Viral Market- Burson-Marsteller%202010%20Global%20 keting: Taking the Temperature of the Hottest ing: Motivations to Forward Online Content.” Social%20Media%20Check-up%20white%20 Trends in Marketing.” Journal of Consumer Mar- Journal of Business Research 63, 9/10 (2010): paper.pdf. keting 25, 3 (2008): 179–182. 1000–1006.

Burt, R. S.  “The Contingent Value of Social Gatignon, H., J. Eliashberg, and T. S. Rob- Horne, C.  “Norm Enforcement in Hetero-

Capital.” Administrative Science Quarterly 42, 2 ertson.  “Modeling Multinational Diffusion geneous Groups: Sanctioning by Majorities and (1997): 339–365. Patterns: An Efficient Methodology.” Marketing Isolated Minorities.” Rationality and Society 20, 2 Science 8, 3 (1989): 231–247. (2008): 147–172.

Burt, R. S.  Brokerage and Closure: An Introduc- tion to Social Capital. New York: Oxford Univer- Gladwell, M.  The Tipping Point: How Little Hovland, C., I. Janis, and H. Kelley.  Com- sity Press, 2005. Things Can Make a Big Difference. New York: munication and Persuasion. New Haven, CT: Yale Back Bay Books, 2000. University, 1953.

Campbell, C., L. F. Pitt, M. Parent, and P.

Berthon.  “Tracking Back-Talk in Consumer- Godes, D., and D. Mayzlin.  “Using Online Kalmijn, M.  “Intermarriage and Homogamy: Generated Advertising.” Journal of Advertising Conversations to Study Word-of-Mouth Com- Causes, Patterns, and Trends.” Annual Review of Research 51, 1 (2011): 224–238. munication.” Marketing Science 23, 4 (2004): Sociology 24, 1 (1998): 395–421. 545–560.

Chiu, H.-C., Y.-C. Hsieh, Y.-H. Kao, and M. Kalyanam, K., S. McIntyre, and J. T.

Lee.  “The Determinants of Email Receivers’ Godes, D., and D. Mayzlin.  “Firm-Created Masonis.  “Adaptive Experimentation in Inter- Disseminating Behaviors on the Internet.” Jour- Word-of-Mouth Communication: Evidence active Marketing: The Case of Viral Marketing nal of Advertising Research 47, 4 (2007): 524–534. from a Field Test.” Marketing Science 28, 4 (2009): at Plaxo.” Journal of Interactive Marketing 21, 3 721–739. (2007): 72–85.

Coleman, J. S.  Foundations of Social Theory.

Cambridge, MA: Harvard University Press, Golan, G. J., and L. Zaidner. “Creative Strat- Kaplan, A. M., and M. Haenlein.  “Users of 1990. egies in Viral Advertising: An Application of the World, Unite! The Challenges and Oppor- Taylor’s Six-Segment Message Strategy Wheel.” tunities of Social Media.” Business Horizons 53,

De Bruyn, A., and G. L. Lilien.  “A Multi- Journal of Computer-Mediated Communication 13, 1 (2010): 59–68. Stage Model of Word-of-Mouth Influence 4 (2008): 959–972. through Viral Marketing.” International Journal Katona, Z. P. P. Zubcsek, and M. Sarvary.  of Research in Marketing 25, 3 (2008): 151–163. Goldenberg, J., S. Han, D. R. Lehmann, and J. “Network Effects and Personal Influences: Dif-

W. Hong. “The Role of Hubs in the Adoption fusion of an Online Social Network.” Journal of

Dobele, A., D. Toleman, and M. Beverland.  Process.” Journal of Marketing 73, 2 (2009): 1–13. Marketing Research 48, 3 (2011): 425–443. “Controlled Infection! Spreading the Brand

Message through Viral Marketing.” Business Granovetter, M. S.  “The Strength of Weak Katz, E., and P. F. Lazarsfeld.  Personal Influ- Horizons 48, 2 (2005): 143–149. Ties.” American Journal of Sociology 78, 6 (1973): ence: The Part Played by People in the Flow of Mass 1360–1380. Communications. Glencoe, IL: Free Press, 1955.

Duan, W. G, B. Gu, and A. B. Whinston. “The

Dynamics of Online Word-of-Mouth and Prod- Greenberg, K..  “New Dunkin’ Donuts Mar- Kiss, C., and M. Bichler.  “Identification of uct Sales—an Empirical Investigation of the keting Runs on Videos.” Marketing Daily, 2008. Influencers—Measuring Influence in Cus- Movie Industry.” Journal of Retailing 84, 2 (2008): Retrieved from http://www.mediapost.com/ tomer Networks.” Decision Support Systems 46, 233–242. publications/article/75137/. 1 (2008): 233–253.

December 2012 JOURNAL OF ADVERTISING RESEARCH 71 Seeding Viral Content

Lawless, J. F., and C. Nadeau.  “Some Simple Niederhoffer, K., R. Mooth, D. Wiesenfeld, Content: Evidence from YouTube.” Information

Robust Methods for the Analysis of Recurrent and J. Gordon.  “The Origin and Impact of Systems Research 23, 1 (2012): 23–41. Events.” Technometrics 37, 2 (1995): 158–168. Cpg New-Product Buzz: Emerging Trends and Implications.” Journal of Advertising Research 47, Trusov, M., A. V. Bodapati, and R. E. Buck- Lee, J., J.-H. Lee, and D. Lee.  “Impacts of 4 (2007): 420–426. lin. “Determining Influential Users in Internet Tie Characteristics on Online Viral Diffusion.” Social Networks.” Journal of Marketing Research Communications of AIS 2009, 24 (2009): 545–556. Phelps, J. E., R. Lewis, L. Mobilio, D. Perry, 47, 4 (2010): 643–658. and N. Raman.  “Viral Marketing or Elec- tronic Word-of-Mouth Advertising: Examining Lin, D. Y., L. J. Wei, I. Yang, and Z. Ying.  Consumer Responses and Motivations to Pass Van den Bulte, C., and Y. V. Joshi.  “New “Semiparametric Regression for the Mean and Along Email.” Journal of Advertising Research 44, Product Diffusion with Influentials and Imita- Rate Functions of Recurrent Events.” Journal of 04 (2004): 333–348. tors.” Marketing Science 26, 3 (2007): 400–421. the Royal Statistical Society: Series B (Statistical Methodology) 62, 4 (2000): 711–730. Phillips, K. W., G. B. Northcraft, and M. A. Van den Bulte, C., and S. Stremersch.  Neale. “Surface-Level Diversity and Decision- Lin, N.  “Social Networks and Status Attain- “Social Contagion and Income Heterogeneity in Making in Groups: When Does Deep-Level ment.” Annual Review of Sociology 25 (1999): New Product Diffusion: A Meta-Analytic Test.” Similarity Help?” Group Processes and Intergroup 467–487. Marketing Science 23, 4 (2004): 530–544. Relations 9, 4 (2006): 467–482.

Lin, N. “Building a Network Theory of Social Portes, A.  “Social Capital: Its Origins and Van den Bulte, C., and S. Wuyts.  Social Net- Capital.” In Social Capital: Theory and Research, Applications in Modern Sociology.” Annual works and Marketing. Cambridge, MA: Market- R. Burt, ed. New York: Aldine de Gruyter, 2001. Review of Sociology 24, (1998): 1–24. ing Science Institute, 2007.

Lin, N, W. M. Ensel, and J. C. Vaughn.  Rogers, E. M.  Diffusion of Innovations. New Wasserman, S., and K. Faust.  Social Network “Social Resources and Strength of Ties: Struc- York: The Free Press, 1962. Analysis: Methods and Applications. Cambridge, tural Factors in Occupational Status Attain- UK: Cambridge University Press, 1994. ment.” American Sociological Review 46, 4 (1981): Salganik, M. J., P. S. Dodds, and D. J. Watts.  393–405. “Experimental Study of Inequality and Unpre- dictability in an Artificial Cultural Market.” Sci- Watts, D. J.  Six Degrees: The Science of a Con- ence 311, 5762 (2006): 854–856. Louch, H.  “Personal Network Integration: nected Age. New York: W. W. Norton & Com- Transitivity and Homophily in Strong-Tie Rela- pany, 2003. Smith, T., J. R. Coyle, E. Lightfoot, and A. tions.” Social Networks 22, 1 (2000): 45–64. Scott.  “Reconsidering Models of Influence: The Relationship between Consumer Social Watts, D. J., and P. S. Dodds.  “Influentials, Mangold, W. G., and D. J. Faulds.  “Social Networks and Word-of-Mouth Effectiveness.” Networks, and Public Opinion Formation.” Media: The New Hybrid Element of the Pro- Journal of Advertising Research 47, 4 (2007): Journal of Consumer Research 34, December motion Mix.” Business Horizons 52, 4 (2009): 387–397. (2007): 441–458. 357–365.

Sun, T., S. Y. G. Wu, and M. Kuntaraporn.  Watts, D. J., and J. Peretti.  “Viral Marketing McPherson, M., L. S. Lovin, and J. M. Cook.  “Online Word-of-Mouth (or Mouse): An Explo- for the Real World.” Harvard Business Review 85, “Birds of a Feather: Homophily in Social Net- ration of Its Antecedents and Consequences.” 5 (2007): 22–23. works.” Annual Review of Sociology 27, 1 (2001): Journal of Computer-Mediated Communication, 11, 415–444. 4 (2006). Retrieved from http://jcmc.indiana.

edu/vol11/issue4/sun.html. Yang, J., C. Yao, W. Ma, and G. Chen.  “A

Newman, A. A.  “Old Spice Argues That Real Study of the Spreading Scheme for Viral Mar-

Men Smell Good.” New York Times, July 16, Susarla, A., J.-H. OH, and Y. Tan.  “Social keting Based on a Complex Network Model.” (2010), B3. Networks and the Diffusion of User-Generated Physica A 389, 4 (2010): 859–870.

72 JOURNAL OF ADVERTISING RESEARCH December 2012