How the characteristics of viral marketer-generated content differ across Twitter and Facebook

(1)

How the characteristics of viral marketer-generated content differ

across Twitter and Facebook

Author: Paula Csatlos

Student number: 11084944 Date of submission: 23 June 2016

Final version

MSc. in Business Administration – Marketing Track Amsterdam Business School, University of Amsterdam

(2)

2

STATEMENT OF ORIGINALITY

This document is written by Student Paula Csatlos who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3 TABLE OF CONTENTS STATEMENT OF ORIGINALITY ... 2 ABSTRACT ... 5 INTRODUCTION ... 6 LITERATURE REVIEW ... 11 Vividness ... 11 Informativeness ... 12 Emotional arousal ... 14 Emotional valence ... 16 Social Media ... 17 Twitter ... 19 Facebook ... 20

Differences between Twitter and Facebook ... 21

Electronic word of mouth ... 22

Viral marketing ... 23 Conceptual framework ... 26 RESEARCH DESIGN ... 28 Research setting ... 28 Data collection ... 29 Variables ... 30 RESULTS ... 33 DISCUSSION ... 39 Theoretical contributions ... 41

(4)

4

Managerial implications ... 42

Limitations and future research ... 43

CONCLUSIONS ... 47

(5)

5

ABSTRACT

What makes certain Facebook and Twitter posts more viral than others? Why do people share a piece of marketer-generated content on social media, but not others? And why do people retweet a post on Twitter, but they do not share a similar post on Facebook? Using data from the Twitter and Facebook pages of the world's most popular five musicians, this paper examines how the vividness, the informativeness and the emotional arousal of generated content shape virality and how the characteristics of viral marketer-generated content differ across Twitter and Facebook. In addition, this paper also aims to contribute to the ongoing debate on whether positive content is more viral than negative content. The results show that vividness increases the likelihood of marketer-generated content to be shared on social media. In addition, non-informative content is more likely to become viral on social media. Finally, there were no conclusive results regarding whether content that evokes high-arousal emotions is more viral than content that evokes low-arousal emotions, nor regarding the way in which the emotional valence of marketer-generated content shapes virality. The results also demonstrate that the social media platform where the content is disseminated moderates the relationship between marketer-generated content and virality. Vivid marketer-generated content is more viral on Facebook than on Twitter. In addition, non-informative content is more likely to become viral on Twitter, whereas the level of informativeness is not a significant predictor of virality on Facebook. These findings contribute to the existing research on viral marketing and social media marketing and shed light on how to design viral marketing campaigns on Facebook and Twitter.

(6)

6

INTRODUCTION

Social media platforms had a meteoric rise since the launch of the first online social platform, Classmates, in 1995, followed by SixDegrees, in 1997 (Piskorski, 2014). Currently, 2 billion people worldwide use social networks such as Twitter, Facebook and LinkedIn, with Facebook alone having 1.5 billion users (“The Economist”, 2015). In the meantime, the global online advertising spending continues to grow, being forecasted to around 150 billion dollars for 2015, and, within that category, the global spending on social-media advertising is forecasted to 20 billion dollars for 2015 (“The Economist”, 2015). As Kim, Sung & Kang explain, “the phenomenal growth of social media has redefined the digital media landscape by changing how information in a networked environment is received and disseminated” (2014, p. 18). Consumers recommend products online, participate in online discussions and share information on social media platforms (Saboo, Kumar, & Ramani, 2015). Thus, sharing online content on social media had become an integrated part of modern life (Berger and Milkman, 2012), with someone tweeting a link to an article published by the New York Times once every four seconds (Harris, 2010).

Sharing online content has a significant impact on consumers and brands alike (Berger et al., 2012). Consumers’ trust in brands and their advertising messages is currently diminishing (Camarero & San José, 2011), and since they are now in control of their experiences on social media and other online environments, consumers choose to consume more user-generated content than marketer generated content (Hoffman et al., 2010). In the meantime, marketers try to disseminate information about their products through electronic word of mouth, instead of traditional communication (Ho & Dempsey, 2010). Whereas the influence of word of mouth on consumers’ beliefs and decisions to purchase increases (Camarero et al., 2011), with 74% of consumers basing their purchase decisions on social media and 43% of social media users sharing the product prior to the purchase (Saboo et al.,

(7)

7 2015), marketers try to leverage the power of consumer-to-consumer communications on social media through viral marketing campaigns (De Bruyn & Lilien, 2008). Consequently, managers and marketers are under constant pressure to motivate consumers to share the marketer-generated content with their friends, in an attempt to influence consumers’ beliefs and purchase decisions (Saboo et al., 2015). In this context, it has become increasingly relevant to examine to what extent the characteristics of marketer-generated content influence the likelihood of the content to be viral. In addition, a deeper understanding of how the social media platform where the content is disseminated moderates the relationship between marketer-generated content and virality could provide valuable guidance for managers and marketers alike.

Nevertheless, while previous literature on viral marketing has focused on its impact on consumer decision making and sales (Goldenberg, Mazursky & Solomon, 2009; Godes & Mayzlin 2004, 2009), little academic research examined the determinants on viral marketing and, the characteristics of viral marketer-generated content, in particular.

Recent research investigated the determinants of virality (Berger et al., 2012; Berger, 2014; Barasch & Berger, 2014), but research on virality in a social media setting is still in an early stage. Berger et al. (2012) focused on the relationship between the level of emotional arousal and virality, Berger (2014) focused on on the motivations of word of mouth generation, Stieglitz & Dang-Xuan (2013) focused on the relationship between emotions and virality, whereas Barasch et al. (2014) examined how the audience size influences the content that people share. However, existing research on viral marketing did not address the unique aspects of information diffusion on social media. Berger et al. (2012) demonstrated that more practically useful content is more viral, and, hence, that informative content is more viral. Nevertheless, their research was focused on what drives people to share online content via email, not on what drives people to share online content on social media. However, what

(8)

8 people share might be moderated by the audience size, with narrowcasting, where people share content to just one person, being different than broadcasting, where people share content to a much larger audience (Berger et al., 2012, Barasch et al. 2014). Turning to the detail, when people are broadcasting, their self-enhancement concerns are heightened, and hence they avoid sharing any type of content that could make them look bad (Barasch et al., 2014). On the contrary, when people are narrowcasting, they tend to focus on others, instead of focusing on themselves, and hence they tend to share content that is useful to the other person (Barasch et al., 2014). Hence, the research of Berger et al. (2012) focused on what drives people to narrowcast, not on what drives people to broadcast, so the findings of Berger et al (2012) that showed that more informative content is more likely to be more viral might not hold in a broadcasting context. In addition, Berger et al. (2012) also investigated how high-arousal emotions contribute to the virality of online content, showing that content that evokes high arousal emotions is more likely to be shared, but, again, their research focused only on narrowcasting. Stieglitz et al. (2013) also examined the relationship between emotions and virality, showing that emotionally charged tweets are more likely to be re-tweeted than neutral tweets. While they did investigate information diffusion in a social media setting, their research examined only the amount of sentiment of each tweet, not the level of emotional arousal evoked by each tweet, making no distinction between high arousal and low arousal emotions. Hence, this paper will draw on the work of Berger et al. (2012) and will examine the impact of the level of informativeness and emotional arousal on virality in a social media setting.

Furthermore, despite the importance of marketer-generated word of mouth on social media for companies, the virality of marketer-generated content on social media has received little attention so fat. The research of Berger et al. (2012) examined the virality of New York Times articles, whereas the context of the research of Stieglitz et al. (2013) was political

(9)

9 communication. In addition, De Vries, Gensler & Leeflang (2012) identified several characteristics of popular marketer-generated content on social media. The authors examined how the level of vividness and the level of informativeness influence the popularity of marketer-generated content on social media, measured by the number of comments and the number of likes of each brand post (De Vries et al., 2012). Nevertheless, to the author’s knowledge, no previous academic research investigated how the level of vividness influences the likelihood of marketer-generated content to become viral.

More importantly, while previous research has studied specific social media platforms in isolation, there is no previous research on viral marketing which incorporated more than one type of social media platform for comparative purposes. Thus, to the author’s knowledge, there is no academic literature on how the characteristics of viral content differ across Twitter and Facebook and, hence, on how social media platform moderates the relationship between marketer-generated content and virality.

This leads to the research question investigated in this paper: To what extent does social media platform moderate the relationship between marketer-generated content and virality? Hence, the research objectives are: (1) to establish the characteristics of viral marketer-generated content in a social media setting, (2) to examine to what extent the level of vividness, informativeness, emotional arousal and emotional valence influence the likelihood of the content to be viral on Twitter and Facebook, and (3) to discover the differences between the two social media platforms regarding the characteristics of viral marketer-generated content.

Based on existing research (Berger et al., 2012; De Vries et al., 2012) and on a preliminary inductive analysis of marketer-generated content on social media by the author, characteristics of viral content have been identified: the vividness, the informativeness and the

(10)

10 emotional arousal of the content. In addition, this paper also aims to contribute to the ongoing debate on how the emotional valence of the content influences virality. Hence, this paper will first examine to what extent the four characteristics of marketer-generated content influence the likelihood of the content to be viral on social media. Secondly, as Twitter, a microblogging site, and Facebook, a social network, represent different forms of social media, with different architecture, norms and culture (Smith, Fischer & Yongjian, 2012), this paper will also examine how the characteristics of viral marketer-generated content differ across the two social media platforms, and, hence, how the social media platform where the content is disseminated moderates the relationship between marketer-generated content and virality.

(11)

11

LITERATURE REVIEW Vividness

The vividness of a marketer-generated post may influence the likelihood of the post to be shared on social media. Vividness is defined as “the representational richness of a mediated environment as defined by its formal features; that is, the way in which an environment presents information to the senses” (Steuer, 1992, p. 81). It consists of two dimensions: the breadth and the depth of the post, with breath being “the number of sensory dimensions simultaneously presented” (Steuer, 1992, p. 81) and depth being “the resolution within each of these perceptual channels” (Steuer, 1992, p. 81). As Nisbett & Ross explain, “information may be described as vivid, that is, as likely to attract and hold our attention and to excite the imagination to the extent that it is (a) emotionally interesting, (b) concrete and imagery-provoking, and (c) proximate in a sensory, temporal, or spatial way” (1980, p. 45). In addition, the degree of vividness depends upon whether dynamic animations, images or contrasting colors have been included in the marketer-generated post (De Vries et al., 2012).

Previous research on vividness shows that the higher the level of vividness, the more enduring and more positive the attitudes towards a website (Coyle & Thorson, 2001). In addition, higher levels of vividness of marketer-generated posts appear to generate a higher number of likes (De Vries et al., 2012).

In addition, videos are considered to be a rich media tool (Coyle et al., 2001) and, since they stimulate two senses, sight and hearing, they are argued to be more vivid than texts or images (De Vries et al., 2012). Considering that on both Twitter and Facebook, posts can contain either text, images or videos, this paper examines to what extent the vividness of marketer-generated content influences the likelihood of the content to be shared on social media and whether this influence is stronger on Twitter than Facebook.

(12)

12 The author expects a positive relationship between the level of vividness of marketer-generated content and virality. More importantly, the author posits that this relationship is moderated by the social media platform where the marketer-generated content is disseminated. Based on a preliminary inductive analysis of marketer-generated content on Facebook and Twitter by the author, it would seem that the format of viral marketer-generated content on Twitter is usually text, while the format of viral marketer-generated content on Facebook is usually video. Hence, the author offers the following hypotheses:

Hypothesis 1a: The higher the level of vividness of marketer-generated content, the more viral the content.

Hypothesis 1b: The positive effect of vividness on virality is larger for Facebook than for Twitter.

Informativeness

The pursuit of information is an important reason why people consume marketer-generated content (Muntinga, Moorman & Smit, 2011). Thus, marketer-marketer-generated posts that contain information about the products, the brand or the company meet the brand's social media followers’ motivations to consume content (De Vries et al., 2012).

Previous research shows that informative ads on social media platforms generate more positive attitudes (Lewin, Strutton & Taylor 2011), whereas Berger et al. (2012) propose that one of the main reasons people share content is that it contains useful information. In addition, their field study demonstrates that more practically useful content is more viral (Berger et al., 2012). Hance, based on these findings, it can be posited that the positive relationship between the level of informativeness and the likelihood of the content to be shared may also apply in a social media setting.

(13)

13 However, as Barasch et al. (2014) showed, when people are narrowcasting, they tend to focus less on themselves and more on others, and hence they tend to share content that is useful to the other person. However, this does not apply to broadcasting, as people focus more on themselves and on self-enhancement when talking to a larger group of people (Barasch et al., 2014). Since the research of Berger et al. (2012) is conducted on narrowcasting, it should be stressed out that there is a chance their findings do not apply in a broadcasting context, such as sharing content on social media. In other words, in a social media setting, informative content may be equally viral or less viral than non-informative content.

Hence, this paper examines to what extent the informativeness of marketer-generated content influences the likelihood of the content to be shared on social media and whether this influence is stronger on Twitter than Facebook. In accordance to the findings of Berger et al. (2012), the author expects a positive relationship between the level of informativeness of marketer-generated content and virality. More importantly, the author posits that this relationship is moderated by the social media platform where the marketer-generated content is disseminated. Considering that informative brand-related user generated content is more common on Twitter than on Facebook, as the former is commonly used to share news or information about brands or other topics (Smith et al., 2012), the author expects viral marketer-generated content on Twitter to be more informative than viral marketer-generated content on Facebook. Hence, the author offers the following hypotheses:

Hypothesis 2a: The higher the level of informativeness of marketer-generated content, the more viral the content.

Hypothesis 2b: The positive effect of informativeness on virality is larger for Twitter than for Facebook.

(14)

14

Emotional arousal

The emotional character of marketer-generated posts may also affect its likelihood to be shared. People discuss with others most of their emotional experiences, while customers generate more word of mouth when they are extremely satisfied or extremely dissatisfied (Berger et al., 2012). In addition, Rimé (2009) claims that both positive and negative emotions drive social sharing. In other words, people are inclined to share emotional experiences independently of the emotional valence of those experiences (Rimé, 2009).

Besides being positive or negative, emotions also evoke different levels of arousal (Smith & Ellsworth, 1985). High arousal emotions can be either positive, such as joy, or negative, such as fear (Heilman, 1997). Similarly, low arousal emotions can be either positive, such as satisfaction, or negative, such as sadness (Heilman, 1997). Arousal is defined by Berger et al. (2012, p. 193) as “a state of mobilization”, with low arousal being characterized by a state of calm and relaxation and high arousal being characterized by activity (Berger et al., 2012). Previous research showed that high arousal emotions increase the likelihood of action related behaviours (Gaertner & Dovidio, 1977; Brooks & Schweitzer, 2011). Considering that sharing content requires action, the activation generated by high arousal emotions should increase the likelihood of content to be shared (Berger et al., 2012). In this case, even if two emotions have the same valence, they may have different effects on dissemination if they trigger different levels of activation (Berger et al., 2012). Previous research showed that content that evokes awe, a high-arousal positive emotion, or anger and anxiety, high-arousal negative emotions, is more viral than content that generates sadness, a low-arousal negative emotion (Berger et al., 2012). Relatedly, Berger (2011) showed that even arousal generated by other factors than emotion, such as running in place, increases the likelihood of content sharing.

(15)

15 Whereas, the research of Berger et al. (2012) is conducted on narrowcasting, based on these findings, it can be posited that the positive relationship between the level of emotional arousal and the likelihood of the content to be shared may also apply in a social media setting. Furthermore, previous research on the link between emotions and virality on social media shows that the sentiment of tweets has an influence on information diffusion on Twitter, in terms of both retweet quantity and retweet speed (Stieglitz et al., 2013). Turning to the detail, tweets that exhibit a large amount of sentiment, either positive or negative, are more likely to be retweeted than neutral tweets (Stieglitz et al., 2013). In addition, emotionally charged tweets are retweeted more quickly than neutral ones (Stieglitz et al., 2013).

Hence, this paper will draw on the work of Berger et al. (2012) and will examine to what extent the level of emotional arousal of marketer-generated content influences the likelihood of the content to be shared on social media and whether this influence is stronger on Twitter than Facebook. The author expects a positive relationship between the level of emotional arousal of marketer-generated content and virality. More importantly, the author posits that the relationship between the level of emotional arousal and virality is moderated by the social media platform where the marketer-generated content is disseminated. Based on a preliminary inductive analysis of marketer-generated content on Facebook and Twitter by the author, it would seem that posts that evoke high-arousal emotions are more likely to be shared on both Facebook and Twitter than posts that evoke low-arousal emotions. Hence, the author offers the following hypotheses:

Hypothesis 3a: The higher the level of emotional arousal of marketer-generated content, the more viral the content.

Hypothesis 3b: The positive effect of emotional arousal on virality is equally large for Facebook and for Twitter.

(16)

16

Emotional valence

In addition to examining how the level of emotional arousal influences virality, this paper is also aiming to contribute to the ongoing debate on whether positive content is more likely to become viral than negative content. Godes et al. (2005) claim that people are more likely to share negative content, but their hypothesis was not tested. Similarly, Stieglitz et al. (2013) examined whether people are more likely to share negative content than positive content on Twitter, but they found no supporting evidence for their hypothesis. In the meantime, Berger et al. (2012) demonstrated that positive content is more viral than negative content. Consumers often create word of mouth out of self-presentation motives (Wojnicki & Godes, 2008; Berger, 2014) or in order to communicate identity (Berger, 2014). As a result, people are more likely to share positive content, as positive content makes the sender look better (Berger, 2014). In addition, people may avoid sharing negative content because they do not want to be associated with negative things or to be perceived as a negative person. However, other research showed that sharing negative content can increase the likelihood of desired impressions, with reviewers being perceived as more competent when they wrote negative reviews, compared to when they wrote positive reviews (Amabile, 1983).

Finally, previous research showed that people share more positive word of mouth when they talk about themselves, in order to make themselves look good, but, in the meantime, they share more negative word of mouth when they talk about others, in order to make themselves look better compared to the others (Berger, 2014). Hence, whether people are talking about their own experiences, or about others’ experiences is influencing the valence of word of mouth (Berger, 2014). When sharing marketer-generated content, people are talking about others, so they might as well be more likely to share more negative marketer-generated content on social media.

(17)

17 Hence, this paper will draw on the work of Berger et al. (2012) and will also examine whether positive content is more likely to become viral than negative content. More importantly, the author posits that the relationship between the valence and virality is moderated by the social media platform where the marketer-generated content is disseminated. Based on a preliminary inductive analysis of marketer-generated content on Facebook and Twitter by the author, it would seem that positive posts are more likely to be shared on both Facebook and Twitter than negative posts. Hence, the author offers the following hypotheses:

Hypothesis 3c: The more positive the marketer-generated content, the more viral the content.

Hypothesis 3d: The positive effect of emotional valence on virality is equally large for Facebook and for Twitter.

Social Media

In recent years, social media platforms received much attention from marketing scholars for their ability to accelerate electronic word of mouth for brands (Kim et al., 2014) and for offering numerous opportunities for firms to use word of mouth on their behalf (Trusov, Bucklin & Pauwels, 2009).

Social media was defined by Kaplan and Haenlein (2010, p. 61) as “a group of Internet-based applications that build on the ideological and technological foundations of Web 2.0, and that allow the creation and exchange of user generated content”. There are various non-exclusive purposes for which a social media platform can be used. Firstly, the social media platform can represent a media outlet, where the marketers disseminate content to consumers (Toubia & Stephen, 2013). Then, it can represent a viral marketing platform, where the firm determines consumers to share marketer-generated content or tracks word of

(18)

18 mouth that occurs organically. Another purpose of a social media platform could be to monitor consumers’ social talk, in order to extract consumer insights (Toubia et al., 2013).

The emergence of social media changed the traditional marketing paradigm, whereby marketers controlled the tools of the promotional mix (Mangold & Faulds, 2009). Turning to the detail, social media has two promotional roles, as it enables both company-to-consumer communication and consumer-to-consumer communication (Mangold et al., 2009). While the company-to-consumer communication is consistent with the traditional paradigm, as marketers still control elements such as the content, timing or frequency of communications, the conversations occurring on social media platforms between consumers limit the control marketers have over the content, timing and frequency of brand-related information (Mangold et al., 2009). Hence, social media allows consumers to be more than passive recipients of marketer-generated content created by marketers. Consumers can now create and disseminate user generated content and actively engage with companies and other consumers to learn from others and share their own experiences (Saboo et al., 2015). As a result, marketing scholars also refer to social media as consumer-generated media (Mangold et al., 2009). Therefore, marketers must learn to transform social media platforms into influence networks, that trigger followers’ interest and generate sales, and to leverage the power of consumer-to-consumer discussions to promote their brands (De Bruyn et al., 2008).

In addition, social media is an umbrella term that describes various types of social media platforms, among which there are microblogging sites, such as Twitter, social networks, such as Facebook, content communities, such as YouTube, collaborative projects, such as Wikipedia, or virtual social worlds, such as Second Life (Kaplan & Haenlein, 2011). This paper focuses on marketer-generated content from two of the most popular social media platforms in terms of marketer interest and engagement of users: Twitter and Facebook (Smith et al., 2012).

(19)

19

Twitter

Twitter is the largest and most popular microblogging site (Jansen, Zhang, Sobel & Chowdury, 2009). The term microblogging comes from the fact that users have Twitter pages, where all their tweets and retweets are aggregated into a single list (Jansen et al., 2009). Nevertheless, while microblogging is specific for microblogging sites, social networks, including Facebook, also have a microblogging feature embedded, represented by status updates (Stieglitz et al., 2013).

Founded in 2006, Twitter allows people to post short text updates that cannot exceed 140-characters or ‘tweets’ to their social networks, reply to or forward tweets (Smith et al., 2012; Marwick, 2011). Besides text, Twitter posts, tweets, may include pictures, videos or hyperlinks. Twitter shares a set of characteristics with all microblogging sites: the text messages are short, the messages are delivered instantaneously and the users need to subscribe to other users in order to receive updates (Jansen et al., 2009). The default setting of Twitter posts is public, which permits Twitter users to follow each other and read the tweets posted by others (Jansen et al., 2009). In contrast to Facebook, that has undirected social networks, when users subscribe to the tweets of other users on Twitter, a practice known as following other users, a directed social network is formed, through which tweets are propagated (Ma, Sun, & Kekre, 2015).

“Tweets commonly ask for or share information, news, opinions, complaints, or details about daily activities” (Smith et al., 2012, p. 103). Hence, Twitter's culture is focused on facilitating conversations and sharing opinions, news and information, whether on brands or on other topics (Smith et al., 2012). In addition, Twitter’s unique feature of forwarding tweets, ‘retweeting’, is a powerful mechanism of information sharing, which makes Twitter

(20)

20 an ideal platform for users to share information to new set of audiences, represented by the followers of the user who retweeted (Stieglitz et al., 2013).

As Jansen et al. (2009) demonstrate, only 19% of tweets are brand-related and, of these, almost half do not have the brand as the primary focus of the tweet. In the cases when brands are central to the post, users express their opinions about the brand and ask or share information about the brand. Nevertheless, due to its 140 character limit, which makes it difficult to address more topics in a single tweet, Twitter hosts more brand-central content than other social media platforms (Smith et al., 2012).

Facebook

Facebook is a social network, founded in 2004 and initially designed for college students only. Users can create Facebook personalized profiles featuring personal details, such as work and education backgrounds and favorite interests (Zywica & Danowski, 2008), and can create a social network by friending other users. Furthermore, users can comment and like other users' posts, write on other users’ walls and follow brands by liking their pages (Smith et al., 2012).

While on Twitter self-presentation occurs through ongoing tweets (Marwick, 2011), on Facebook, it occurs through both content and personal profiles (Zywica et al., 2008). Furthermore, Facebook also allows users to use brands in their self-presentation, by means of listing their education background or favorite sports teams, even though the culture of self-promotion through brand referencing is not highly developed on this social media platform (Smith et al., 2012).

Furthermore, marketers can create brand pages on Facebook, where they can disseminate information about new products and the brand and, thus, provide conversation starters for their followers, in addition to providing an environment where those

(21)

consumer-to-21 consumer conversations can happen. In addition, Facebook allows its users to post on brand pages a similar content to the marketer-generated one (Smith et al., 2012).

Differences between Twitter and Facebook

While both Twitter and Facebook are user centric, which implies that they are characterized by high levels of social interaction (Ma et al., 2015), the two represent different forms of social media platforms and each of them has a distinctive architecture and a unique culture (Smith et al., 2012).

Due to its 140 character limit, which makes it difficult to address more topics in a single tweet, Twitter hosts more brand-central content than Facebook does (Smith et al., 2012). Additionally, due to Twitter’s cultural focus on sharing information, Twitter brand-related posts are more likely to be informative than Facebook posts (Smith et al., 2012). In addition, compared to Twitter, Facebook is more oriented towards enabling social connectedness for its users, as it “allows people to build or maintain social capital, communicate with others, keep up with other peoples’ lives, and discover rumors and gossip” (Smith et al., 2012, p. 103).

Finally, Facebook offers more options for self-presentation than Twitter. On Twitter, self-presentation occurs through ongoing tweets, rather than personal profiles, and blatant self-promotion is often considered inappropriate (Marwick, 2011). On Facebook, on the other hand, self-presentation occurs through both content and personal profiles (Zywica et al., 2008). In addition, Facebook also allows users to use brands in their self-presentation (Smith et al., 2012).

(22)

22

Electronic word of mouth

Social media platforms serve as an ideal venue for brand-related electronic word of mouth, a venue where consumers ask for or disseminate information and opinions about products or brands to their social networks (Kim et al., 2014). The product-related or brand-related information that users share on social media platforms is more likely to be perceived as trustworthy and credible than other forms of electronic word of mouth (Kim et al., 2014). In addition, individuals voluntarily expose themselves to brand-related content on social media platforms, by clicking on the ‘like’ button on Facebook or on the ‘follow’ button on Twitter (Kim et al., 2014).

Word of mouth (WOM), defined by Westbrook (1987, p. 261) as “informal communications directed at other consumers about the ownership, usage, or characteristics of particular goods and services or their sellers”, was found to increase product awareness (Van den Bulte & Wuyts, 2009) and to influence the adoption of new product categories and the choice of specific brands within existing categories (East, Hammond & Lomax, 2008), as consumers rely on word of mouth in order to reduce both the uncertainty and the perceived risk of the purchase decision (Murray, 1991). With over three billion brand impressions generated through social talk on a daily basis (Keller & Libai, 2009), word of mouth influences consumer behaviour from the movies they choose to watch, to the internet websites they access (Chevalier & Mayzlin, 2006; Godes & Mayzlin, 2009). A recent study by Bughin, Doogan & Vetvik (2010) revealed that word of mouth is the trigger of 20 to 50 per cent of all decisions to purchase, and, compared to paid advertising, generates more than twice the sales.

While it is clear that social talk is frequent and has a huge impact on consumers and consumer behaviour, less is known about what drives people generate and share word of mouth. In an extensive review, Berger argued that “word of mouth can be understood in terms

(23)

23 of five key functions that it serves for the word of mouth transmitter: impression-management, emotion regulation, information acquisition, social bonding, and persuasion” (2014, p. 588). Hence, people share word of mouth in order to communicate who they are, to express desired identities and for self-enhancement purposes. Another reason people share word of mouth is to manage their emotions, the way in which they experience these emotions and the way in which they express these emotions (Berger, 2014). Third, people share word of mouth in order to seek information from others, by bringing up the topic of interest (Berger, 2014). Fourth, sharing word of mouth helps at connecting with others, as well as at strengthening social connections. Finally, people share word of mouth in order to persuade others (Berger, 2014).

With the evolution of internet and social media, a new form of word of mouth has emerged: the electronic word of mouth (eWOM), defined by Camarero et al. (2011, p. 2293) as “any informal communication using IT concerning the usefulness of certain goods or services, as well as sellers or suppliers”. Electronic word of mouth has several advantages, compared to traditional word of mouth. The first advantage refers to the higher reach and diffusion speed of pieces of content on electronic word of mouth. Whilst the reach of traditional word of mouth is restricted to the size of the social network of each individual, the electronic word of mouth reaches a much wider audience (Kaplan et al., 2011). Secondly, compared to traditional word of mouth, electronic word of mouth is easier to monitor, thus the impact of electronic word of mouth on sales or profit can be better analyzed (Kaplan et al., 2011).

Viral marketing

Viral marketing is regarded as a form of word of mouth advertising, in which consumers communicate with each other about brands (Camarero et al., 2011). The term was

(24)

24 first coined in 1996, by Jeffrey Rayport, who explained the exponential growth pattern of viral marketing by comparing the diffusion of content with the spread of viruses (Kaplan et al., 2011). Viral marketing was defined by Kaplan et al. (2011, p. 255) as “electronic word-of mouth whereby some form of marketing message related to a company, brand, or product is transmitted in an exponentially growing way, often through the use of social media applications”. Hence, it relies on the creation of provocative content to incentivize customer-to-customer communication of branded content (Camarero et al., 2011). Social media is particularly suited for viral marketing, as the community element of social media increases the convenience of sharing the piece of content to a wide audience. Thus, viral marketing is sometimes referred to by researchers as social media marketing (Kaplan et al., 2011).

Viral marketing allows marketers to leverage consumer-to-consumer communications to promote information about products or services in a cost effective manner, generating similar levels of awareness that were previously achievable only with television advertising and generating a fastest adoption by the market (De Bruyn et al., 2008; Kaplan et al., 2011). Viral marketing’s driver is the effectiveness of online recommendations to increase awareness, trigger interest and incentivize product purchase (De Bruyn et al., 2008). In addition, a key advantage of viral marketing is the distribution of marketer-generated content to others and the exponential growth that results subsequently (Camarero et al., 2011). Hence, knowing what drives people to engage in viral marketing is extremely relevant. This leads to the question raised by Berger et al. (2012, p. 192), “Is virality just random, as some argue (e.g., Cashmore 2009), or might certain characteristics predict whether content will be highly shared?”.

Previous research points to several determinants of viral marketing success, including the characteristics of content (Porter & Golan, 2006; Berger et al., 2012), the structure of the social network (Bampo et al., 2008), the characteristics of the recipients (Hennig-Thurau et

(25)

25 al., 2004; Camarero et al., 2011; Phelps et al., 2004), the role of influential users (Iyengar, Van den Bulte et al., 2011), and the seeding strategy (Bampo et al., 2008; Libai, Muller & Peres, 2005; Kalish, Mahajan & Muller, 1995). Turning to the detail, the research on content characteristics focused on content attractiveness (Porter et al., 2006), content usefulness (Pousttchi & Wiedemann, 2007) and the emotional aspects of the content (Berger et al., 2012). The research on social network structure focused on the characteristics of the social network where the content is disseminated, such as the size and the structure of the network (Bampo et al., 2008). The research on the characteristics of the recipients of the content as determinants of viral marketing focused on the behavioral, motivational and social characteristics of the recipients (Camarero et al., 2011; Bampo et al., 2008; Hennig-Thurau et al., 2004). The research on the seeding strategy focused on the minimum social network size required to generate viral marketing (Bampo et al., 2008; Libai et al., 2005). Finally, the research on the role of influence in the viral process focused on the role of influental users, who have strong social network presence or ties (Iyengar et al., 2011).

The work of Berger et al. (2012) is the most relevant for this paper. Their findings indicate that the valence of the content affects its virality, since positive content was found to be more viral than the negative one. The authors went beyond the valence of the content to examine how the activation induced by specific emotions drives content sharing, and discovered that content that generates awe, a high-arousal positive emotion, or anger and anxiety, high-arousal negative emotions, is more viral than content that generates low-arousal emotions. Hence, this paper will draw on the work of Berger et al. (2012) and will focus on informativeness, emotional arousal and emotional valence as determinants of viral marketing on Twitter and on Facebook. In addition, based on existing research (De Vries et al., 2012) and on a preliminary inductive analysis of marketer-generated content on social media by the author, an additional determinant of virality has been added: vividness.

(26)

26

Conceptual framework

A graphical illustration for the conceptual framework for the determinants of the virality of marketer-generated content is presented in Figure 1.

Figure 1. The moderating role of the social media platform on the relationship between marketer-generated content and virality

More specifically, this paper investigates the virality of marketer-generated content based on three determinants: the level of vividness, the level of informativeness and the level of emotional arousal of the marketer generated-content. The author expects a positive relationship between the characteristics of marketer-generated content and virality. An additional determinant of virality of marketer-generated content to be examined in this paper is the emotional valence of the marketer-generated content.

More importantly, the author posits that this relationship is moderated by the social media platform where the marketer-generated content is disseminated. Turning to the detail, the author expects the postitive effect of vividness on virality to be larger for Facebook than Vividness

Emotional arousal

Virality Informativeness

Social media platform plaplatform

(27)

27 for Twitter. The positive effect of informativeness on virality is expected to be larger for Twiter than for Facebook. Finally, both the positive effect of emotional arousal on virality and the positive effect of emotional valence on virality are expected to be equally large for Facebook and for Twitter.

(28)

28

RESEARCH DESIGN Research setting

The music industry is used as the context of this research. The importance of the music industry on social media is shown by the fact that seven of the top ten most followed Twitter accounts are musicians (Saboo et al., 2015).

The music industry is “characterized by music artists (brands) offering their music (products) to listeners (consumers)” (Saboo et al., 2015, p. 2). While brands have been traditionally associated with companies, products or services, marketing scholars introduce the term human brands to refer to to any “well-known persona who is the subject of marketing communication efforts” (Thomson, 2006, p. 104). The musicians that are at the centre of the music industry are excellent examples of human brands.

Unlike products brands that cannot engage with consumers, human brands can both directly interact and create emotional bonds with consumers (Saboo et al., 2015). The rise of social media platforms, such as Facebook and Twitter, has enhanced the level of engagement between musicians and their listeners, by providing a venue where consumers can connect with their brands. Indeed, many musicians now use social media to interact with their listeners. Furthermore, the typical social media website of a musician “contains all the elements of a self-contained marketing eco-system” (Saboo et al., 2015, p. 2), as it contains branding elements, provides information on the musician, streams promotional videos and is a platform for free sampling (Saboo et al., 2015). In addition, the social media website represents the perfect venue where consumers can express their social identity and exert social influence on others, while providing marketers with quantitative and qualitative insightful feedback (Saboo et al., 2015).

(29)

29 Additionally, Aggarwal & McGill (2012) found that consumers consider brands as less relevant than humans due to their inanimate nature. Previous academic research also shows that humanizing brands that were originally inanimate generates positive customer attitudes and, hence, improves the performance of these brands (Puzakova et al., 2013). The humanization of brands is primarily done through conversations with consumers on social media platforms (Gensler, Völckner, Liu-Thompkins & Wiertz, 2013). Hence, understanding how human brands interact with customers and what characteristics of their social media content generate virality is important since humanized brands could replicate these best practices in their attempt to engage with customers.

Hence, the present research uses data from the Twitter and Facebook pages of the world's most popular five musicians, in other words the musicians that have the largest number of followers on both Twitter and Facebook: Taylor Swift, Justin Bieber, Katy Perry, Rihanna, and Selena Gomez.

Data collection

To answer the research question, to what extent does social media platform moderate the relationship between marketer-generated content and virality, a quantitative analysis was undertaken. Hence, the data was collected in a standardized way (Saunders & Lewis, 2012), through social media research. Data was collected from the Twitter and Facebook pages of the world’s most popular five musicians, with the aid of R, a software for statistical computing that mines data using the APIs (Application Programming Interfaces) of Twitter and Facebook. Furthermore, RStudio was used as a user interface for R.

The author chose to study Twitter and Facebook not only because they are the most popular social media platforms in terms of marketer interest and number of users (Smith et al.,

(30)

30 2012), but also because Twitter’s feature of ‘retweeting’ and Facebook’s feature of ‘sharing’ are powerful mechanisms of content diffusion.

To keep the data collection manageable, yet capture a representative data set, the paper examines the marketer-generated posts of five human brands, over a 17-week period, from December 24, 2015 to April 21, 2016. For each Twitter or Facebook post, the content of the post, the format of the post, the number of retweets or shares, respectively, and the sentiment classification were observed. Duplicate posts were ignored and non-English posts were filtered out.

Variables

Virality. The dependent variable in this research is the virality of content. To measure the virality of marketer-generated content, the number of shares on Facebook and the number of retweets on Twitter will be considered. Opinion-passing is an enhanced form of electronic word of mouth on social media platforms (Kim et al., 2014) and, in order to be viral, marketer-generated content needs to be passed-on by brands’ followers to their friends. Since retweeting is the opinion-passing mechanism on Twitter and sharing is the opinion-passing mechanism on Facebook, the two have been chosen as a measure for virality of marketer-generated content. The number of shares of each Facebook post and the number of retweets of each Twitter post have been computed as a percentage of the total number of the followers of the Facebook or Twitter page, respectively. Then, both the number of shares and the number of retweets have then been standardized.

Vividness. The first independent variable in this research is the level of vividness of the marketer-generated content. To measure vividness, the format of the post will be taken into account. Posts containing videos will be considered as having high levels of vividness, posts containing photos will be considered as having medium-high levels of vividness, posts

(31)

31 containing links will be considered as having medium-low levels of vividness, whereas posts containing text will be considered as having low levels of vividness. Dummy variables have been created for video, photo and text, whereas link has been chosen as the baseline group.

Informativeness. The second independent variable in this research is the level of informativeness of the posts. To measure informativeness, the number of words in each post will be counted, then standardized. Long posts will be considered as being more informative, whereas short posts will be considered as being less informative.

Emotional arousal. The third independent variable in this research is the level of emotional arousal evoked by the content of the post. The measurement of the level of emotional arousal will be done with the aid of a sentiment analysis tool, Sentiment. The sentiment analysis tool classifies the emotions of Facebook and Twitter posts as surprise, joy, fear, disgust, sadness, anger, and unknown. In addition, in order to test the reliability of the sentiment analysis tool, the author manually coded a subset of 10% of the total number of Facebook and Twitter posts. The manual coding of the subset of social media posts was significantly and positively correlated with the automated coding. Hence, it has been demonstrated that the sentiment analysis tool is reliable. In addition, according to Heilman (1997), anger, fear, joy and surprise are high-arousal emotions, while sadness is a low-arousal emotion. In addition, according to Lang, Greenwald, Bradley & Hamm (1993), disgust is a high-arousal emotion. Hence, the Twitter and Facebook posts that evoke surprise, joy, fear, disgust and anger will be considered as having a high level of arousal. Similarly, the posts that evoke sadness will be considered as having a low level of arousal. Dummy variables have been created for surprise, joy, fear, disgust, sadness, anger, whereas unknown has been chosen as the baseline group.

(32)

32 Emotional valence. In addition, the emotional valence of the Facebook and Twitter posts has also been taken into account. The coding of the valence has been done with the aid of a second sentiment analysis tool, Sentiment140, that classified the posts as positive, negative, or neutral. Similarly, in order to test the reliability of the second sentiment analysis tool, the author manually coded a subset of 10% of the total number of Facebook and Twitter posts. The manual coding of the subset of social media posts was significantly and positively correlated with the automated coding. Hence, it has been demonstrated that the second sentiment analysis tool is also reliable. Dummy variables have been created for positive and negative polarities, whereas neutral polarity has been chosen as the baseline group.

Social media platform. The moderator in this research is the social media platform where the marketer-generated content is disseminated. Two types of social media platforms are examined in this research: a microblogging site, Twitter, and a social network, Facebook. As Twitter and Facebook represent different forms of social media, with different architecture, norms and culture (Smith et al., 2012), it is posited that the social media platform moderates the relationship between marketer-generated content and virality. A dummy variable for the social media platform has been created, with twitter posts coded as 1 and Facebook posts coded as 0.

(33)

33

RESULTS

The author examined the relationship between virality, content characteristics and social media platform where content is disseminated using regression, with the number of shares or tweets, as the dependent variable, and the number of characters, the dummy variables for the type of content, the dummy variables for emotions, and the dummy variables for emotional valence, as independent variables. In addition, in order to test for the moderating effect of the social media platform where the content is disseminated on the relationship between marketer-generated content and virality, the author created interactions for all independent variables with the moderator, where Twitter is 1 and Facebook is 0. The regression analysis found that the content characteristics and the social media platform where the content is disseminated influence the virality of content (F(23, 728) = 4.025, p < 0.01). In addition, the model explained 11.3% of variance in virality. The results of the regression are presented in Table 1.

Coefficient B Std. Error Beta t Sig.

Number of characters 0,00 0,06 - 0,004 - 0,07 0,95

Dummy negative valence 0,06 0,32 0,01 0,19 0,85

Dummy positive valence 0,08 0,16 0,03 0,50 0,62

Dummy joy - 0,22 0,18 - 0,08 - 1,23 0,22

Dummy surprise - 0,66 0,56 - 0,06 - 1,17 0,24

Dummy anger - 0,28 0,68 - 0,04 - 0,42 0,68

(34)

34

Dummy sadness - 0,32 0,18 - 0,08 - 1,76 0,08

Dummy video 0,69 0,13 0,25 5,20 0,00

Dummy photo 0,42 0,10 0,20 4,08 0,00

Dummy text - 0,12 0,56 - 0,05 - 0,22 0,83

Dummy positive valence Twitter - 0,29 0,20 - 0,10 - 1,44 0,15 Dummy negative valence Twitter - 0,13 0,49 - 0,01 - 0,27 0,79

Number of characters Twitter - 0,15 0,08 - 0,11 - 1,96 0,05

Dummy joy Twitter 0,40 0,22 0,12 1,83 0,07

Dummy surprise Twitter 0,36 0,74 0,03 0,49 0,63

Dummy anger Twitter 0,05 0,74 0,01 0,07 0,94

Dummy fear Twitter 1,64 1,08 0,12 1,52 0,13

Dummy disgust Twitter - 0,79 0,68 - 0,04 - 1,17 0,24

Dummy sadness Twitter 0,38 0,30 0,06 1,28 0,20

Dummy video Twitter - 0,12 0,20 - 0,02 - 0,59 0,56

Dummy photo Twitter 0,39 0,13 0,13 2,95 0,00

Dummy text Twitter 0,66 0,56 0,27 1,18 0,24

Table 1. Regression with virality as a dependent variable

Hypothesis 1a. This hypothesis predicted that vivid content is more likely to be viral than content with low levels of vividness. In addition to being significantly and positively

(35)

35 correlated to virality, the dummy variable for video (β = 0.25, p < 0.05) and the dummy variable for photo (β = 0.20, p < 0.05) are significant predictors of virality on Facebook. As expected, posts containing a video or photo have a high level of vividness, which, in turn, increases the likelihood of the post to be shared on Facebook. However, the dummy variable for text (β = – 0.05, NS) is not significant on Facebook. Similarly, the dummy variable for photo (β = 0.13, p < 0.05) is a significant predictor of virality on Twitter. As expected, posts containing a photo have a high level of vividness, which, in turn, increases the likelihood of the post to be shared on Twitter. However, the dummy variable for video (β = – 0.02, NS) and the dummy variable for text (β = 0.27, NS) are not significant on Twitter. Vivid content is more likely to be viral on both Facebook and Twitter, whereas the types of content associated with a low level of vividness are not significant predictors of vividness on either Facebook or Twitter. Hence, Hypothesis 1a is supported.

Hypothesis 1b. This hypothesis predicted that the relationship between vividness and virality is more likely to be stronger on Facebook, than Twitter. The dummy variable for video (β = 0.25, p < 0.05) and the dummy variable for photo (β = 0.20, p < 0.05) on Facebook recorded a higher Beta value than the dummy variable for photo (β = 0.13, p < 0.05) on Twitter. The positive effect of vividness on virality is larger for Facebook than for Twitter. Hence, Hypothesis 1b is supported.

Hypothesis 2a. This hypothesis predicted that informative content is more likely to be shared than content with low levels of informativeness. In addition to being significantly and negatively correlated to virality, the number of characters is not significant (β = – 0.004, NS) on Facebook and a significant predictor of virality on Twitter (β = – 0.11, p = 0.05). Contrary to expectations, informative posts are less likely to be shared than posts with low levels of inormativeness on Twitter. Hence, Hypothesis 2a is not supported.

(36)

36 Hypothesis 2b. This hypothesis predicted that the positive effect of informativeness on virality is larger for Twitter than for Facebook. The level of informativeness is not a significant predictor of virality on Facebook. Meanwhile, whereas the effect on informativeness on virality is indeed larger for Twitter than for Facebook, contrary to expectations, informative posts are less likely to be shared on Twitter. Hence, Hypothesis 2b is not supported.

Hypothesis 3a. This hypothesis predicted that content that evokes high arousal emotions is more likely to be shared. The dummy variable for fear is significantly and positively correlated to virality and the dummy variable for sadness is significantly and negatively correlated to virality. In addition, the dummy variable for sadness (β = – 0.08, p < 0.1) is marginally significant on Facebook. As expected, posts that evoke sadness, a low-arousal emotion, are less likely to be shared on Facebook. However, the dummy variables for joy (β = – 0.08, NS), surprise (β = – 0.06, NS), anger (β = – 0.04, NS), and fear (β = – 0.08, NS) were not significant on Facebook. Similarly, the dummy variable for joy (β = 0.12, p < 0.1) is marginally significant on Twitter. As expected, posts that evoke joy, a high-arousal emotion are more likely to be shared on Twitter. However, the dummy variables for surprise (β = 0.03, NS), anger (β = 0.01, NS), disgust (β = – 0.04, NS), fear (β = 0.12, NS), and sadness (β = 0.06, NS), were not significant on Twitter. Despite the fact that posts that evoke low-arousal emotions are less viral and posts that evoke high-arousal emotions are more viral, the majority of the findings have been inconclusive. Hence, Hypothesis 3a is not supported.

Hypothesis 3b. This hypothesis predicted that the positive effect of emotional arousal on virality is equally large for Facebook and for Twitter. On Facebook, only the dummy variable for sadness (β = – 0.08, p < 0.1) is marginally significant. On Twitter, on the other hand, only the dummy variable for joy (β = 0.12, p < 0.1) is marginally significant. Hence,

(37)

37 due to the fact that the majority of the findings have been inconclusive, Hypothesis 3b is not supported.

Hypothesis 3c. This hypothesis predicted that positive content is more likely to be shared than negative content. However, neither the dummy variable for positive valence (β = 0.03, NS), nor the dummy variable for negative valence (β = 0.01, NS) are significant on Facebook. Similarly, neither the dummy variable for positive valence (β = – 0.10, NS), nor the dummy variable for negative valence (β = – 0.01, NS) are significant on Twitter. Hence, Hypothesis 3c is not supported.

Hypothesis 3d. This hypothesis predicted that the positive effect of valence on virality is equally large for Facebook than for Twitter. The dummy variable for positive valence and the dummy variable for negative valence are not significant for either Facebook or Twitter. Hence, Hypothesis 3d is not supported.

The results of the research conducted in this paper are summarised in Table 2.

Hypothesis Results

Hypothesis 1a The higher the level of vividness of

marketer-generated content, the more viral the content.

Supported

Hypothesis 1b The positive effect of vividness on virality is larger for

Facebook than for Twitter.

Supported

Hypothesis 2a The higher the level of informativeness of

Not supported

Hypothesis 2b The positive effect of informativeness on virality is

larger for Twitter than for Facebook.

(38)

38

Hypothesis 3a The higher the level of emotional arousal of

Not supported

Hypothesis 3b The positive effect of emotional arousal on virality is

equally large for Facebook and for Twitter.

Not supported

Hypothesis 3c The more positive the marketer-generated content, the