Can valence and arousal alone predict eWOM behaviour on SNS?

Can Valence and Arousal Alone Predict eWOM Behaviour on SNS? Reina Anne Q. Regis

Graduate School of Communication, University of Amsterdam

Author’s Note

Reina Anne Q. Regis, Graduate School of Communication, University of Amsterdam This paper was supervised by Lotte Van Willemsen, PhD. Correspondence concerning this thesis should be addressed to Reina Anne Regis, Contact: reinaanne.acad@live.com.ph

Abstract

Social network sites (SNS) provide brands the innovation to spread marketing information to target audiences for the purpose of raising brand awareness, building positive brand attitudes, and reaching potential customers, through electronic word-of-mouth (eWOM). The goal, therefore, is to facilitate eWOM by stimulating SNS users to forward marketing content. The motivations behind what induces customers to perform eWOM behaviour (share, like, or comment on brand posts), however, remains a question. This paper attempts to investigate this issue through the mechanisms of emotion sharing, taking into account how valence (positive vs. negative) differ in that respect, and examined the potential role of physiological arousal (high vs. low) in driving eWOM. Evidence suggest eWOM behaviour is not only driven by emotional arousal. A few control variables were found to correlate highly with eWOM behaviour, but did not confound in the relationship with valence and arousal. Limitations and suggestions for future research are discussed.

Keywords: eWOM marketing, SNS marketing, social network sites, emotion sharing, social sharing

Can Valence and Arousal Alone Predict eWOM Behaviour on SNS?

Social network sites (henceforth: eWOM) offer brands and consumers, alike, to converge and interact in profitable ways. For instance, the interactive environment of SNS (e.g. Facebook) provide businesses with a channel to broadcast marketing information to a wider audience simultaneously (Svensson, 2011). However, not only businesses possess the ability to take advantage of this interactive environment. Now, more than ever, consumers possess increased ability over the creation and spread of marketing information as either user-generated or shared commercial content (Cho, Huh, & Faber, 2014). User-user-generated content could be stories about experiences, with goods and services, with a multitude of other consumers, through electronic word-of-mouth (henceforth: eWOM) (Hennig-Thurau,

Gwinner, Walsh, & Gremler, 2004). This ability of consumers to propagate brand information can actually be profitable for business by way of stimulating eWOM to get people to share advertisements and create user-generated content (e.g. about brands, products, and services) (Kaplan & Haelein, 2010).

In the current context, eWOM is defined as ‘any positive or negative statements made by potential, actual, or former customers about a product or brand, and which is made

available to a multitude of people and institutions via the Internet (Henning-Thurau et al., 2004). Hence, eWOM can garner a variety of positive outcomes for brands when consumers start sharing advertisements. Indeed, research shows eWOM affects brand attitudes and brand awareness (Berger & Milkman, 2012). Additionally, it impacts people’s choices, information diffusion, and sales (Berger & Iyengar, 2013).

To get the full advantage of eWOM, brands must encourage individuals to share the information to their social networks. Previous research on the subject of sharing on social

media and eWOM found highly ‘emotional content’ are more frequently shared. Emotional content is the capacity of a message (e.g. content) to elicit sentimental arousal on the message recipient (Hansen et al., 2011), and consists of a valence and an arousal dimension. While valence refers to whether a content displays positive or negative emotions (Berger & Milkman, 2009), arousal is a heightened physiological activation or action or an excitatory state (Berger & Milkman, 2009), which has been shown to encourage performance of action-related behaviours. For example, anger and sadness are negative emotions but differ in the level of intensity they evoke – that is, anger involves activation while sadness, deactivation.

With respect to valence, Berger’s and Milkman’s (2009) study of New York Times articles found that positive news were more likely to make it to the most-emailed list than negative content. A plausible explanation for such findings is that rarity of positive content in news type information makes it break through the clutter, making readers notice them more. The basis of such assumption is derived from the baseline context proposed by Wu (2012) who found that in a negative baseline, positive information was more noticeable and

supported over negative information. To relate this finding to that of Berger and Milkman’s (2012) findings, positive content in the news could be seen as novel in a negative baseline online newspaper environment, and therefore readers see this novelty as shareworthy.

However, in the none-news type context, such as SNS advertising, a negativity effect might be expected. That is, people may be more likely to put a heavier weight on the negative information and share it more than ads with positive content. The negativity effect is defined as the greater weighing of negative as compared with equally extreme positive information in the formation of judgments (Ahluwalia, 2002, p. 270). Because this subject has not been previously investigated, results of the present study would be highly imperative to the field of communication science. Findings will not only be of value for theory building on the

marketing strategies. To address this void in the literature, the first aim of this study is to examine the sharing of emotional (positive vs. negative) content in the context of marketing communication.

The second aim of the study is to examine the effects of emotional valence in relation to arousal. In that respect, the current study draws upon the finding of Berger and Milkman (2012) that the effects of valence is overridden by the effect of arousal. More specifically, they found that highly arousing emotion, such as anger and fear, are more likely to incite sharing behaviour than low arousing emotions (e.g. sadness, contentment). The authors explain this finding by the nature of arousal to cause physiological activation. The

mobilization brought about by this excitatory state may boost information sharing (Berger, 2011). The question, however, is whether similar effects are to be expected when arousal is measured as a driver of eWOM. Indeed, Berger and Milkman (2009) found a positive

relationship between arousal and information sharing upon investigation on the issue. That is, emotionally arousing stories are the ones more frequently shared.

The robustness of the said finding, nevertheless, remains under examination

considering that online news platforms and SNS environments are two different worlds. This is because editorial decisions about newsworthiness hinge on stories perceived to appeal to the public’s imagination (Bennett & Daniel, 2002). Thus, stories about tragedies ‘fit the bill’ because they are ‘dramatic, emotive, and awe-inspiring’ (p. 34). However, contents posted within the SNS environs does not go through such ‘gatekeeping’ (i.e. a social scientific concept that Lewin conceived in 1947 to describe the story-selection process of the news media (Clayman & Reisner, 1998)) because users are free to post user-generated content (Svenssion, 2011) with the motivation to put oneself in a positive light e.g. for

self-enhancement (Berger & Iyengar, 2013). In other words, SNS environments have a positive baseline.

It is therefore imperative to examine whether the principles of the negativity effect and the mechanisms that drive individuals to take action following arousal from an emotionally intense negative source would apply to eWOM marketing. Such that, arousal from negative emotions positively correlates with individuals’ intention to share negative highly arousing brand posts. By applying the mechanisms of arousal as described by Taylor (1991) and Cannon (1932, cited by Taylor, 1991) and the principles of the negativity effect in positive baseline SNS marketing environment, as discussed above, the predicted direction is that negative high arousing brand posts could evoke the performance of eWOM behaviour more, while a lesser effect would be observed for low arousal negative and positive brand posts.

Theoretical background Marketing on social network sites

SNSs have attracted a huge global following over the last recent years since its inception. To put the said matter into perspective, 74 percent of at least two billion online users use social network sites (“Social Networking Fact Sheet,” n.d., para. 1). Ellison (2007) defines SNS as ‘Web platforms where individuals can create profile in which privacy can be set to semi-/public, invite a limited number of users whom they share connections with, and navigate information created and shared by others within their/and extended networks within the system’ (p. 211).

From a marketing perspective, SNS have offered businesses with lucrative

opportunities to communicate marketing messages (Trusov et al., 2009). To capitalize on this opportunity, many businesses, nowadays, are concentrating much of their advertising efforts online, particularly, on SNS (Soares et al., 2012). This advertising campaigns spending is projected to amount to $23.68 billion by the end of the year 2015 (“Social network ad spending”, 2014, April 15, para. 1).

However, given the fact that the average SNS user encounters, on average, 1,000 commercial messages a day, it has become difficult for marketers to break through the clutter (Gritten, 2007). Yet, clutter is only one of the issues marketers’ encounter with regards to users’ negative response to advertisements. One pressing issue is the users’ general dislike of Internet ads (Cho, 2002). Consequently, it was found that participants were very likely to avoid ads they saw on SNS (Kelly, Kerr, & Drennan, 2010)

Although there is evidence to suggest that consumers can be passive to SNS ads, Campbell, Ferraro, and Sands (2014) decided to explore deeper into whether all consumers are, indeed, passive. The researchers then identified five segments of SNS users, namely: passive (29%), talkers (28%), hesitant (24%), active (10%), and averse (9%). From these findings it can be concluded that passive users actually comprise only of about a third of SNS users. Interestingly, the said findings found that 28 percent are ‘talkers’, those who are likely to engage in eWOM behaviors such as brand engagement, referrals, and willingness to share information.

Their findings shows potential for marketers to increase ad effectiveness on social media. The less stellar findings about consumer reaction to social media advertisements, therefore, does not warrant a total abandonment of using this medium for marketing. In fact, Gritten (2007) suggested that marketers should change with the changing marketing

landscape, adopting strategies consumers would respond positively to. This means changing the advertising approach on social network sites – one that not only attracts attention but also encourages passive users to actively participate in online social transmission on SNS.

A good advertising approach would then be one that coincides with consumer

behaviour on the internet (Clemons, Barnett, & Appadurai, 2007). What do SNS users prefer doing on a typical day of browsing on Facebook? A 2014 survey revealed users enjoy using the status updates feature (“What and why do people share on social networks?” Adweek,

July 15, 2014) which allows them to post or re-post content. The same report found that 49 percent of people share marketing information about products hoping to ‘change opinions or encourage actions’.

The advantages of eWOM marketing

The social sharing of marketing information, electronically, is referred to as eWOM; and on the social network site, Facebook, eWOM constitutes sharing of, liking, and

commenting on brand information. This sharing of marketing information has several advantages for both businesses and consumers. As a matter of fact, these advantages have been documented as early as the 1950’s when Katz and Lazarsfeld (1955) suggested that traditional WOM has a major influence on consumers’ buying decisions and choice of media content exposure.

Nowadays, the technological advancements on SNS have allowed users to openly manifest their attitudes about brand and product information, and disseminate it to beyond the local community, making for a faster dissemination with a much wider reach (Wu, Wang, & Hsu, 2014). To examine the marketing advantages of eWOM, Chevalier and Mayzlin (2006) examined the link between Amazon.com and bn.com (the Barnes and Nobles Web site) customer reviews and book sales. Indeed, they found a positive link between the number of reviews of a book and the average star rating (number of stars), and amount of sales within the study period.

In addition to stimulating the intention to purchase, the effectiveness of eWOM as a marketing strategy has also demonstrated to reverse consumers’ negative attitude towards brands and products. This is because source credibility is higher when consumers receive positive reviews about brands/products from their personal relations than when the

information derives from advertisers (Cho, Ho, & Faber, 2014; Gunawan & Huarng, 2015). Therefore, an effective SNS marketing strategy would be one that engages SNS users to

spread eWOM. The current study examines two strategies that could potentially lead people to share content: valence and arousal. It draws upon the literature about the influence of emotional valence and arousal on people’s sharing behaviour.

In as much as the present paper would like to give a definition of ‘emotion’, authors are still at a disagreement as to what constitutes emotions, and whether to define it through a dimensional approach or discrete feelings (Botha & Reyneke, 2013; Heilmann, 1998;

Mulligan & Scherer, 2012). Therefore, for the current context, emotion will be given a dimensional approach using a two-factor model proposed by Tellegen and Watson (1985). The two-factor model describes emotions as ranging from intensity (high to low) and direction (positive to negative) (Watson & Tellegen, 1985).

Arousal and intensity, which comprise emotions according to the literature reviewed above, are factors that increased the incidence of social transmission. How emotions

stimulate people to take action will be discussed in the succeeding literature. Rime, Corsini, and Herbette (2002), for instance, posited that social transmission of emotions occur because of people’s inherent need to share emotions with others. A principle echoed by Berger and Milkman (2014) who found that people, indeed, talk over their emotional experiences with others.

The motivation to share certain types of emotions with others is believed to have varying underlying motivations. People share emotions with others to make sense of their own experience, reduce emotional dissonance (i.e. the conflict between expressed and experienced emotions (Abraham, 2000)), or deepen social connections (Berger & Milkman, 2012). Sharing of emotions could also be motivated by people’s need to be part of a social group, to be individualist or express personality, or/and the need for personal growth (Ho & Dempsey, 2010), and driven by self-enhancement (Berger & Iyengar, 2013). Therefore, emotions are integral in the individuals’ information sharing behaviour.

While emotionally-charged information, in general, has impact on a social forwarding behaviour, the size of its impact may depend on its valence (positive vs. negative).

Specifically, evidences suggest that people prefer forwarding positive content (Berger & Milkman, 2012; Botha & Reyneke, 2013; Guadagno, Rempala, Murphy, & Okdie, 2013).

Berger and Milkman (2012) who, by far, conducted the largest analysis aimed at tackling the psychological mechanisms behind why people share emotionally-charged information with others. In their study, the researchers analysed around 7,000 unique New York Times articles published on the website of New York Times (i.e., nytimes.com) to track which types of stories people emailed the most. The researchers found, even with varying subjects and controlling for other possible confounding variables (e.g. placement on website), a strong link between positive stories and sharing behaviour.

The said findings – although contradictory to the classic theory of diffusion on mass media which argue that negative information has more ‘news value’ (Hansen, Arvidsson, Nielsen, Colleoni, & Etter, 2010) – could be explained by the fact that the pervasiveness of negative news (on nytimes.com) made people get accustomed to them. In other words, the rarity of positive stories makes them break through the clutter and, therefore, receive more notice. Another explanation for why people prefer sharing positive emotions is that the conveyance of positivity builds the impression of being someone who would rather make others feel better than upset or angry (Berger & Milkman, 2012).

In a similar vein, Botha and Reyneke (2013) in their research explored the relationship between virality and emotions. They used video stimuli and conducted in-depth interviews with university students from Generation Y (born between 1978 and 1992). Findings showed people who responded positively to the videos expressed higher likelihood to share the video to their peers than those whose responses were negative.

Further evidence to Internet users’ preference to share positive content is provided in the study of Guadagno, Rempala, Murphy, and Okdie (2013). The said researchers

investigated the role of emotional response to a video source in the likelihood to spread an internet video. Results showed that the participants who felt positively about the videos they watched expressed strong preference to pass it on than those who felt negatively about the video. However, participants also expressed higher likelihood to pass on video that evoked negative emotions over neutral videos.

Altogether, these findings suggest that online sharing behaviour is largely influenced by emotions – people prefer sharing positive emotional content over the negative emotional content. However, might these findings apply to eWOM marketing in the context of SNS? The robustness of the aforementioned findings (Berger & Milkman, 2012; Botha & Reyneke, 2013; Guadagno et al., 2013) have not actually been tested in in the social network marketing context. What is known is that when consumers are faced with both positive and negative marketing information the latter is given more attention. This is known as the ‘negativity effect’, characterised as the greater weighting of negative information compared with equal amount of positive information when forming overall evaluations about an object (Ahluwalia, 2002).

The negativity effect has been explained by its diagnostic character. Negative

information is perceived to be highly diagnostic, and categorizes a stimulus object more (Wu, 2013; Ahluwalia, 2002). That is because when compared with positive or neutral information negative information is perceived more useful as it brings about aversion to the risks

associated with a stimulus (Herr, Kardes, & Kim, 1991). While the negativity bias is well-documented in the psychology literature (Ahluwalia, 2002), it does not unequivocally apply to marketing context of eWOM. To investigate the negativity effect, Wu (2013) performed a large-scale analyses of Amazon.com reviews to examine whether negativity effects are

contingent of the context. Wu investigated this phenomenon by a series of experiments in which he compared positive and negative reviews. Findings from two experiments showed no significant difference in the perceived helpfulness of positive and negative information. Wu went on to deduce that the negativity effect is a rather ‘context-dependent’ phenomenon. Wu suggested that in real life, negative events occur less often, giving people a ‘positive baseline’ (i.e., reference point). A positive baseline makes negative occurrences look rarer and gravitates people’s attention towards them.

To test this assumption, Wu performed a third experiment in which people exposed to positive versus negative overall review evaluations as baseline conditions, in combination with either positive or negative individual reviews. Wu’s findings suggested that context indeed explains the presence of a negativity effect. That is, participants’ attention were directed towards either positive or negative information when that type of information appeared rarer when in relation to the baseline valence. Thus when the majority of the

consumers have evaluated a product positively, an individual reviews gets more weight when judging the value of that review when it negative in valence and vice versa. It makes

information stand out more, which increases the diagnostic character of the review.

In line with the principles of the negativity effect, the positive baseline of marketing communication on SNS, and in the context of sharing behaviour on SNS, the expectation is that consumers exposed to a negative valence brand post will have a higher willingness to share, like, or comment on the post than participants exposed to a positive valence brand post (H1).

The effect of arousal on eWOM behavior

The current study also investigated on the potential role of arousal in predicting eWOM behaviour (to share, like, and/or comment). This hypothesis was derived from earlier evidences showing that physiological arousal stimulated information-forwarding behaviour

(Berger, 2011; Berger & Milkman, 2012). To describe the rationale of the current

investigation physiological arousal has first to be defined. Arousal, as a physiological state, refers to the activation of the autonomic system (Heilman, 1997), “marked by the secretion of catecholamines, leading to increases in heart rate, blood pressure, blood sugar, and respiration (Taylor, 1991, p. 68). According to Heilman, an aroused organism is alert and ready to take action, whereas, an unaroused organism is lethargic and unavailable to process stimuli. Examples of intense emotions evoking physiological arousal are awe, anxiety, and fear; while emotions associated with physiological deactivation or low arousal are sadness or anxiety (Heilman, 1997).

To find out how physiological arousal relates to information forwarding behaviour, Berger (2011) conducted an experiment to study the effects of physiological arousal on people’s information sharing. He suggested that, because physiological arousal causes

alertness in individuals leading them to take action, inducing physiological arousal through an emotional content might boost information-sharing behaviour. Indeed, this hypothesis was supported by his own findings whereby individuals made to do a physical activity expressed increased willingness to share a piece of content.

Another evidence on the positive link between physiological arousal and information forwarding behaviour is from Berger and Milkman (2012), in a New York Times study investigating which type of news people emailed most. The researchers observed that while e.g. contentment and awe are both positive emotions, awe-inspiring content were shared more frequently than content evoking the feeling of contentment. For the negative emotions, anger and sadness, people shared content that incited anger over sadness. Berger and Milkman then concluded that high emotional intensities incited physiological arousal may play a role in the online information forwarding behaviour, which was subsequently examined in a second experiment. Participants in the experimental condition were instructed to jog in place for one

minute. After which, they were asked to read an article unrelated to the study, and instructed to forward the content to anyone should they please. Consistent with their expectations that physiological arousal induced information sharing, participants in the experimental condition shared the content more than those who were not asked to do a physical exercise. Therefore, Berger and Milkman suggested that, indeed, physiological arousal arising from high intensity emotion could stimulate people to take action, in which case, forwarding the content to the people they know.

In a similar study aiming to establish a causal relation linking emotional intensity and social sharing had revealed consistent results as the Berger and Milkman experiment. Rimé, Corsini, and Herbette (2002) distinguished the effects of three varying emotional intensities on social sharing behaviour by subjecting participants to three different 5-minute movie excerpts that evoked high, moderate, and low emotional intensities. After watching the films individuals were led to a waiting room and left there for five minutes with a friend. Within the five minutes, conversations were tape recorded without the participants’ knowledge. Analysis into the conversations revealed that the extent of sharing the emotion positively correlated with the intensity of the emotion felt by the participants. That is, the more intense the emotion felt, the more a participant shared the emotion. Rimé et al. (2002) also found that “emotion sharing is a repetitive process that involves several recipients” (p. 187). Furthermore, emotion sharing, as they found, was not limited to the direct friend inside the experiment room but also involved several other persons.

The findings above show that high intensity emotions associated with physiological arousal (e.g. anger, surprise, joy, and fear) likewise induce action-related behaviours such as eWOM (Heilman, 1997). Why does physiological arousal, or arousal, trigger action and low arousal less so? Physiological arousal causes an individual to be awake, alert, thus, increasing preparedness to process stimuli, and act upon it; whereas, an unaroused individual, in contrast,

is lethargic with lower preparedness to process and act on the stimuli (Heilman, 1997). Berger (2011) posits that active behaviour caused by arousal may boost social transmission.

Therefore, in line with the mechanisms of how high emotional intensity evokes arousal consequently leading to emotion sharing and information forwarding behaviour, there is evidence to predict that consumers exposed to highly arousing brand posts would be more willing to share, like, and comment on that post than on the low arousal brand posts (H2). Interaction effect of arousal and valence.

The present study also examined the interaction effects of valence and arousal on stimulating eWOM behaviour. The rationale behind such examination is earlier literature (e.g. Taylor, 1991) that analysed how positive and negative events incited varying emotional intensities leading to physiological arousal. Such finding led to the assumption that the interaction of valence and arousal may, indeed, produce differing results with regards to eWOM behaviour given that it has been established how physiological arousal boosted social forwarding behaviour (Berger & Milkman, 2012; Berger, 2011; Rimé et al., 2002). On the role of valence, in an aroused state, on eWOM behaviour will be discussed in the succeeding paragraphs.

The earliest evidence to the differential effects of valence in evoking physiological and behavioural responses was first described by Walter Cannon in 1932 (cited in Taylor, 1991) as the flight-or-fight reaction. Cannon’s proposal was derived from the “frameworks that examined arousal and its correlates” (p. 68). That is, the implicit assumption is tha positive events do not evoke as much emotional intensity as negative events, suggesting therefore that negative events and physiological arousal have a more defined positive link than positive events and physiological arousal (Taylor, 1991).

To apply this principle in the context of emotional content on SNS, might it be that negative, high arousing information (e.g. awe and anger) could induce more intense

physiological arousal in individuals than low arousing negative and positive posts? And could the physiological arousal incited by a negative arousing emotion stimulate the performance of eWOM behaviour? In arriving at a hypothesis, the current study builds upon the mechanisms leading individuals to share emotions felt from exposure to a negative high arousing content. Therefore, the expectations is consumers exposed to negative high arousing brand posts are more willing to share, like, and comment on that post than participants exposed to positive high arousing brand posts. These differences are less pronounced between low arousing negative and positive brand posts (H3).

Methods Design and participants

The present research employs a 2 (valence: positive vs. negative) X 2 (arousal: high vs. low) between subjects, experimental design involving 136 adult Facebook users. However, only 121 finished the survey, and were retained for analysis. Participants come from the

Philippines (66.9 percent), Europe (26.6 percent) including Belgium, Bulgaria, France, Germany, Greece, Italy, The Netherlands, Romania, and Turkey; rest of Asia (3.2 percent) including Japan, China, and India; and Peru (0.8 percent). Of the 121 participants, 59.5

percent were women, and the mean age of all participants is 34.94 (SD=8.87, min = 22; max = 62). The majority of the participants were highly educated: 53.7 percent possessed bachelor’s degrees, 27.3 have masters’ degrees, 15.7 finished some college degree, 1.7 have PhDs, and 1.7 finished high school education.

Stimuli

The stimuli were a simulation of Facebook posts one would see on the UNICEF fan page. UNICEF was chosen as the source of the post because it was considered a gender neutral organization whose advertisements are typically laden with emotions. Therefore, it offers the necessary realistic aspect necessary for the experiment which used Facebook brand

posts. The post contained a two-sentence message told by a fictional UNICEF photographer about a real-life story of a Yemeni girl who was married off to an older man, but mustered the courage to file for divorce that was actually granted in court. The story was identical across conditions except on the part where the emotion was mentioned: “Isn’t it [emotion] how Yemeni girls, like Nojoud Ali are denied a childhood and education? Thanks to the UNICEF and the collective support of local journalists, the Yemeni people are beginning to take a stand against child marriages,” and differed only in the words used to induce the target emotions, namely, anger, sadness, awe, and satisfaction.

Valence – defined as a subjective feeling of pleasantness (positive) or unpleasantness (negative) (Barrett, 1998) – was manipulated by asking participants how i.e. infuriating/sad (negative emotions) or jubilating/nice (positive emotions) it is that Yemeni girls can now go back to school. Arousal was manipulated by constructing the messages so that they evoke feelings of anger and awe (high arousal) and satisfaction and sadness (low arousal). Synonymous words were chosen to avoid making the stimuli and measures too obvious to participants as doing so might prompt the realisation about the real purpose of the current study. The message came along with the UNICEF logo (exactly as it looks like and is positioned on the fan page) and a photo of a young Yemeni girl, in her traditional ‘hijab’ (modest clothing for Muslim women that covers the hair, neck, and entire body). Her whole face can be seen in which the girl was showing a neutral expression. This was selected to avoid the facial expression inducing unintended target emotions.

The four variations were pretested among 20 subjects, and showed participants were able to identify valence significantly after running an ANOVA: valence, F(1,13)=7.084, p=.021. As intended, the post with the positive emotion was perceived as positive (M=5.86, SD=.69) than the post with the negative emotions (M=3.86, SD=1.86). ANOVA results also showed that t the highly arousing brand post was perceived to be more arousing (M=5.33,

SD=1.41) than the low arousing brand post (M=4.30, SD=1.29), however these differences were not significant, F(1,12)=1.053, p=.327, probably due to small sample size.

Procedure

Participants were recruited from the author’s own Facebook and LinkedIn networks via convenience sampling. Survey invitations were sent via private message and as status updates on Facebook and LinkedIn between 9 and 13 July 2015. A few of the author’s personal contacts likewise shared a status update about the study within their own networks.

In the invitation, participants were informed the online experiment is for the

completion of a master’s thesis, hence, all data would be used for research purposes only. As a cover story, they were simply informed the research is about Facebook posts. The invitation mail and status updates contained links to the online experimental page on Qualtrics.

After clicking, participants saw the welcome message stating the purpose of the

research and what it was about (cover story). A declaration of consent was also obtained – and on that page participants could select whether to give or not give consent to participate. Upon clicking on ‘next’ at the end of the consent page, random assignment ensues assigning

participants to one of the four experimental conditions. Thereafter, questions were ordered for all participants as follows: eWOM measures, manipulation questions, identification of emotions incited in the post, involvement with non-profit organizations and UNICEF, message attitude, and demographics.

Measures

Because central to the current study is to investigate people’s sharing behaviour on Facebook, the measures for eWOM were framed within the Facebook context. That is, respondents were asked to rate on a 7-point likert scale (1= “not at all”, 7= ”extremely likely”) (De Vries, Gensler, & Leeflang, 2012) about their willingness to share (M = 4.03, SD = 2.20; min=1,

max=7), like (M = 4.64, SD = 2.10; min=1, max=7), and comment (M = 3.31, SD = 2.07; min=1, max=7) on the post

As manipulation checks for valence and arousal, respondents were asked “How does this story make you feel?” followed by four semantic bipolar pairs (1=low; 7=high). For valence the item was ‘negative feeling – positive feeling’; and for arousal: ‘very passive’-‘very active’; passive’-‘very mellow’-passive’-‘very intense’; passive’-‘very low energy’-passive’-‘very high energy’) (Berger, 2011). Factor analysis showed the three items for arousal fall into one dimension with an eigenvalue of 2.63 explaining 87.78 percent of the total variance. All items appeared to have good internal consistency, ⍺ = .929, and no items needed deleting as it would have reduced the alpha. Items for arousal were then averaged to form a composite index for arousal and make it viable for statistical analysis using the compute variable function to get the mean. Finally, participants were asked to answer a series of questions measuring potential covariates. Specifically, they were asked whether they (0=no; 1 = yes) (1) financially supported non-profit organisations within the past year, (2) financially supported UNICEF within the past year, (3) have been a fan of the UNICEF Facebook fan page within the past year, (4) visited to the UNICEF Facebook page within the past year, which were meant to measure involvement with issue, non-profit organizations in general, and UNICEF in particular. It was deemed imperative to measure source and issue involvement as it has been found to predict the kind of processing involved during an encounter with a persuasive message (Lane, Miller, Brown, & Vilar, 2013)

Attitude towards the message (measured on a 7-point scale) was also included as a control variable because it has been found to be a significant determinant when predicting whether or not an individual will perform the target behaviour (Lane, Miller, Brown, & Vilar, 2014). The said variable was measured with the question: ‘How informative did you think was the post?’ rated on a 7-point (1=low, 7=high) bipolar scale (Cho, Huh, & Faber, 2014).

Age, sex, and education were standard pertinent information for demographics purposes, which were also measured for their potentially confounding effects on the dependent variables.

Results Manipulation and confound check

To check the success of the manipulation, a one-way ANOVA was performed with valence as the independent variable and the manipulation item (‘how does this story make you feel: positive or negative?’) as the dependent variable. Results showed participants were not able to identify the differences between the negative (M=4.34, SD=1.84), and positive valence of the posts (M=4.72, SD=1.89), F(1,109)=1.160, p=.284.

A separate one-way ANOVA was conducted to test success of the arousal

manipulation, with the arousal index as the dependent variable and arousal (high vs. low) as the independent variable. However, participants were unable to differentiate the high (M=3.93, SD=1.34) and low conditions (M=4.28, SD=1.48) resulting in a non-significant differences between conditions, F(1,114)=1.73, p=.19. To conclude, manipulation did not work despite that the pre-test showed a certain significant success (see Discussion).

Financially supported non-profit organisations, financially supported UNICEF, UNICEF Facebook page fan, visited the UNICEF Facebook page, attitude towards the message, age, sex, education, and region (a dummy variable that was assigned the values “1” for all Asian countries, and “0” for all Western countries (see Discussion)) were checked for confounding effects on the dependent variables willingness to share, like, and comment on the Facebook post. A bivariate correlation was conducted as a confound check. The analysis found the following to correlate significantly with the dependent variables: the willingness to share the post was significantly correlated with being a Facebook fan of the UNICEF page, r=.234, p=.013; prior visits to the Facebook page, r=.187, p=.047; attitude towards the

message, r=.616, p<.001; and age r=.193, p=.040; region, r=.380, p<.001. Willingness to ‘like’ the post had a strong positive correlation with attitude towards the message, r=.582, p<.001 and region, r=.255, p=.006. Willingness to comment on the post correlated

significantly with being a Facebook fan, r=.200, p=.034; prior visits to UNICEF’s Facebook page, r=.207, p=.028; attitude towards the message, r=.397, p<.001, and age r=.253, p=.007. Therefore, the said control variables were held constant when the hypothesis were tested. The effect of valence

H1 stated that consumers exposed to a negative valence brand post will have a higher willingness to share, like, or comment on the post than participants exposed to a positive valence brand post. The said dependent variables were tested separately on a univariate ANOVA, with valence as the independent variable. Covariates (see ‘controlling for alternative explanations’) were also incorporated into the analyses.

Willingness to share. The main effect of valence on willingness to share was found to be non-significant; negative valence (M=4.26, SD=2.13), positive valence (M=3.79,

SD=2.27), F(1,113)=.81, p=.371, _{ suggesting group mean differences did not differ}

significantly across conditions. Only the controls – message attitude (F(1,113)=67.31, p<.001); region, F(1,111)=18.378, p<.001, 

=.142; and age, F(1,113)=4.80, p=.030—were found to be highly significant. Yet, none of the correlating controls had any significant confounding effects on participants’ willingness to share the post.

Willingness to ‘like’. The main effect of valence on participants’ willingness to ‘like’ or the post did not result in significant mean differences between the groups. Participants assigned to view the negative message (M=4.63, SD=2.09) rated willingness to ‘like’ the post just about as much as those exposed to the positive message (M=4.66, SD=2.13),

highly significant in this analysis, message attitude F(1,114)=53.31, p<.001, .338; and F(1,111)=7.605, p=.007, 

=.064

Willingness to comment. Exposure to either the negative (M=3.50, SD=2.06) or positive (M=3.11, SD=2.07) post also did not significantly influence participants’ willingness to comment on the posts they saw, F(1,111)=.661, p=.418, =.006. However, differences in means between groups in terms of participants’ message attitude and age were statistically highly significant, F(1,111)=16.396, p<.001, 

=.129 for message attitude, and F(1,111)=6.927, p=.010, 

=.059 for age.

Findings for valence did not support the prediction for H1 because no significant effects of valence on eWOM behaviour were observed. Therefore, H1 is rejected. The effect of arousal

H2 stated that consumers exposed to highly arousing brand posts would be more willing to share, like, and comment on that post low arousal brand posts. Three univariate ANOVA was used to analyse differences in group means between arousal and the dependent variables willingness to share, like, and comment on the post.

Willingness to share. No significant differences in group means were found for low (M=3.98, SD=2.14) and high (M=4.09, SD=2.28) arousal, F(1,113)=.000, p=.988, =.000, while keeping the controls constant. The covariate message attitude, however was a highly significant predictor, F(1,113)=67.35, p<.001, 

=.373, as were age and willingness to share F(1,113)=4.927, p=.028, 

=.042. However, none of the significant results influenced people’s behaviour to share the post. Being a fan of and visit to the UNICEF Facebook page within the past year also did not have significant confounding effects on people’s sharing behaviour.

Willingness to ‘like’. No significant statistical differences in means were found in the levels of arousal on individuals’ willingness to ‘like’ the post. Participants exposed to the low

arousal conditions (M=4.59, SD=2.06) rated their willingness to ‘like’ the post just as much as those who saw the high arousal (M=4.70, SD=2.186) posts, F(1,111)=.002, p=.965, 

=,000. The effects of arousal on message attitude, however, was highly significant F(1,111)=54.861, p<.001, =.331.

Willingness to comment. Participants’ willingness to comment on the post was also not significantly affected by differences in the arousal conditions, as can be seen in the results which show low arousal to have a mean of 3.21 (SD=1.95) compared to the high arousal mean of 3.42 (SD=2.20), F(1,111)=.284, p=.595, 

=.003. Analysis into the group mean differences for message attitude, F(1,111)=16.307, p<.001, 

=.128, and age F(1,111)=6.945, p=.010, 

=.059 showed highly significant results.

The findings above are inconsistent with the expectations of H2. Therefore, H2 has to be rejected for lacking evidential support.

Interaction effects of valence and arousal

H3 stated that consumers who are exposed to negative highly arousing brand posts would be more willing to share, like, and comment on that post than participants exposed to positive highly arousing brand posts; and these differences are less pronounced between low arousing negative and positive brand posts. To test H3, three independent ANOVA were conducted with valence and arousal as fixed factors, and the dependent variables: willingness to share, like, and comment. Controls that significantly correlated with each were also added as covariates (see ‘controlling for alternative explanations’).

Willingness to share. The interaction effects of valence and arousal produced no statistical differences in group means: anger (M=4.31, SD=2.21), sadness (M=4.06, SD=2.04), awe (M=3.79, SD=2.32), contentment (M=3.79, SD=2.25), F(1,109)=2.182, p=.143, 

=.020. Although in the expected direction, these differences were not significant, however. Attitude

towards the message was highly significant: F(1,109)=65.59, p<.001, =.376; and so was age F(1,109)=5.04, p=.024, 

=.044

Willingness to ‘like’. Analysis into the difference in the means of participants’ willingness to ‘like’ the post also did not produce significant results, although differences means were in the expected directions with anger having the highest mean with M=4.76, SD=2.22 compared with sadness (M=4.53, SD=2.02); whereas awe (M=4.66, SD=2.19), and contentment (M=4.66, SD=2.11) were equal, F(1,112)=2.199, p=.141, =.019. Message attitude was the only covariate controlled for in this analysis which resulted in a highly significant difference, F(1,112)=60.476, p<.001, 

=.351.

Willingness to comment. Group means also did not differ statistically: anger (M=3.58, SD=2.12), sadness (M=3.44, SD=2.05),  (M=3.28, SD=2.30), low positive (M=2.93,

SD=1.82), F(1,109)=.128, p=.721, =.001. A highly significant result was found for message attitude F(1,109)=16.01, p<.001, _{=.128, and age F(1,109)=6.793, p=.010, }

=.059.

The findings above describe no significant interaction between valence and arousal on eWOM behaviour, and do not support expectations of H3. Therefore, H3 is rejected.

Discussion

What drives SNS users to share content with their online social network? The current study builds on from past findings on social transmission online and the influence of emotions in the sharing behaviour. The study was aimed at investigating the individual and combined effects of valence and arousal in social transmission in the context of the social network site, Facebook.

In the first hypothesis, the expectations was that consumers exposed to a negative valence brand post will have a higher willingness to share, like, or comment on the post than participants exposed to a positive valence brand post. This was drawn with respect to the well-documented negativity effect (Wu, 2013; Ahluwalia, 2002) which holds that negative

information is given more weight than positive information because the former is perceived to be more diagnostic and useful in assessing a stimulus object. Therefore, participants exposed to negative marketing message (posts inciting anger and sadness) would be more willing to share, like, and comment on the post more than positive content (posts inciting awe and satisfaction).

The second hypothesis stated that consumers exposed to highly arousing brand posts would be more willing to share, like, and comment on that post than for low arousal brand posts. That is, that high arousing messages (the ones intended to evoke anger and awe) are expected to increase individuals’ willingness to share, like, and comment on the high arousing messages in a post than on the low arousing messages (satisfaction and sadness) in a post. This expectation is in line with the principle which holds that physiological activation associated with arousal drives individuals to take the specified behavioural response (Heilman, 1997; Berger, 2011); in which case is eWOM behaviour in the current context.

Contradictory to these predictions, findings from the current experiment showed no support for these two main effects (i.e. effects of valence and arousal on eWOM behaviour). That is, negative and positive valence, as well as, low and high arousal did not differ in stimulating eWOM behaviour on Facebook (i.e. willingness to share, like, and comment on the posts).

In addition to examining the individual effect, this study also looked at the combined effects of valence and arousal. Hypothesis 3 stated that consumers who are exposed to negative highly arousing brand posts would be more willing to share, like, and comment on that post than participants exposed to positive highly arousing brand posts; and that these differences are less pronounced between low arousing negative and positive brand posts. The interaction effects of valence and arousal – with the expectation that the negative and high arousing condition would be more willingly shared, liked, and comment on than low/negative

messages, high/positive messages, and low/positive messages – also did not yield significant differences in eWOM behaviour. This finding did not support the hypothesized expectation (H3). The non-significant findings suggest that eWOM behaviour on SNS (i.e. Facebook) may not be entirely driven by message valence and arousal alone. Several theoretical explanations for the non-significant findings are discussed below.

First, as Wu (2012) suggested, the negativity effect may not always apply in eWOM behaviour contexts. This suggestion was based on his own finding in which negative and positive information were rated equally in terms of helpfulness when the quality of information was controlled for. In other words, negative information may draw people’s attention to its novelty, but when presented with a comparably informative positive

information the effect of negative information on diagnosticity is attenuated (Wu, 2012). In relation to the findings of the current study, Wu’s finding gives insight into the

non-significant difference in effects of both positive and negative information on eWOM

behaviour in this way: the post messages in the four variations informs about a real life story of a child bride, which level of informativeness was formulated equal. Consequently,

consistent with Wu’s finding but not the current expectation, the effects of positive and negative valence on eWOM behaviour produced no significant difference.

In a similar vein, findings from an even earlier study investigating the role of the negativity effect on perceived diagnosticity (Ahluwalia, 2002) had suggested that the effectual attenuation of negative information when compared with an equally extreme positive

information was moderated by brand familiarity. That is, individuals significantly paid more attention to positive information of familiar than unfamiliar brands, and more so when it was consistent with pre-existing attitudes (Ahluwalia, 2002). Existing attitudes about UNICEF, a name that has existed for decades, however, were not considered in the current study but which could have contributed to the non-significant difference in effects of negative and

positive information on eWOM behaviour. Therefore, future research could investigate the negativity effect on familiar brands to test how it compares with positive information, and its consequent effects on diagnosticity.

A second theoretical explanation that could explain the non-significant results is individual characteristics which may form a boundary condition. This is such that certain personality traits including openness, extroversion, and a sense of self-image (Kabadayi & Price, 2014) could influence people’s motivations to perform eWOM behaviour (share, like, and comment on a post). Individuals’ expression of personality as how they want others to perceive them was also found as a key factor influencing the types of emotions in a message that people may want to share (Ho & Dempsey, 2010). Potentially, people may only share, like and comment upon branded FB posts based on valence and arousal, when they have a natural tendency to engage in such behaviours given personality characteristics. In follow-up research it is recommended to test for such individual differences.

Another explanation for the non-significant findings is how online users are differentially segmented, so that the ‘talkers’ (those who actually rate high on brand

engagement and eWOM) comprise only of about 28% (Campbell et al.., 2014). This means that the remaining 72 percent of SNS users just may not be wont to forward information. Future research could take into account user segmentation as a mediator of the relationship between valence, arousal, and eWOM behaviour.

The material used in the current study could also be a factor. The issue of child brides is considered a gross human rights violation in many societies that do not practice it. In fact, the Universal Declaration of Human Rights (1948) explicitly states that child marriage seriously affects girls’ and women’s rights to health, education, equality, non-discrimination, and to live free from violence and exploitation (“What is the impact of child marriage: Human Rights and Injustice,” n.d., para.1). This means that despite the marketing goals of the

UNICEF Facebook fan page, the issue on child brides in a post message could still be

adjudged as negative news which according to Berger and Milkman (2012) are less frequently shared than positive news in the news segment. Hence, subsequent research could use

marketing messages that do not contain topics others could be religiously and morally sensitive to.

In relation to the topic, cultural differences in eWOM behaviour was considered. However, the inclusion of region distinguishing Asian and Western participants in the main analysis produced no significant outcome for eWOM behaviour. Therefore, cultural

differences between Asian and Western participants in relation to the effects of valence and arousal on Facebook eWOM behaviour could not have been a plausible explanation.

Nevertheless, it deserves closer scrutiny due to the significant difference in participants’ responses with regards to willingness to perform eWOM behaviour. This could be explained by Hofstede’s (1980) cultural dimension model which differentiates tightly- and loosely-knit social frameworks, the former with a preference to look after the family in exchange for unquestioning loyalty while the latter has a preference to look after of the self (Hofstede, 1980). These differences could have implications for marketers who want to study people’s motivations to perform eWOM behaviour, which type of content Asians and Westerns like to share on social networks, as well as, the prevalence of eWOM behaviour being practiced in each. Future research interested in comparative studies across countries could therefore take cultural differences into consideration using a larger sample, or conduct a content analysis of eWOM within each country of interest.

Finally, vividness – as indicated by the richness of a post’s formal features meant to stimulate the different senses (De Vries, Gensler, & Leeflang, 2012) – is also a factor that could explain the non-significant results of the current study. De Vries et al. found that by

achieving creating a dynamics of colours, animation, and pictures in a post liking was enhanced consequently boosting liking and commenting behaviour.

In contrast, the image used in the stimuli (see Appendix) was one with a dark

background, and the girl wearing the usual hijab in black (typical conservative clothing worn by Muslim women that covers the hair and entire body). The low resolution image with a measurement of 480-640 pixels – which is considered low compared to high resolution 3000x2400 pixel images (“Change the size or resolution of a picture,” n.d., para. 1) – may also have contributed to the low quality of the image. In that case, with reference to De Vries et al. (2012) description on what constitutes vividness, the image in the current study’s stimuli appeared bleak. Therefore, the image did not have the characteristics that would invoke liking and commenting on.

References

Social networking fact sheet (n.d.). In Pew Research Center. Retrieved from: http://www.pewinternet.org/fact-sheets/social-networking-fact-sheet/

Social network ad spending to hit $23.68 Billion worldwide in 2015: Advertisers in North America spend the most to be social”, April 15, 2015, EMarketer, Retrieved from: http://www.emarketer.com/Article/Social-Network-Ad-Spending-Hit-2368-Billion-Worldwide-2015/1012357

Abraham, R. (2000). The role of job control as a moderator of emotional dissonance and emotional intelligence–outcome relationships. The Journal of Psychology, 134(2), 169-184.

Ahluwalia, R. (2002). How prevalent is the negativity effect in consumer environments? Journal of Consumer Research, 29(2), 270-279.

Berger, J. & Iyengar, R. (2013). Communication channels and Word of Mouth: How the medium shapes the message. Journal of Consumer Research, 40(3), 567-579. Berger, J. (2011). Arousal increases social transmission of information. Psychological

Science, 22(7), 891-893.

Berger, J., & Milkman, K. (2012). Social transmission, emotion, and the virality of online content. Wharton Research Paper.

Botha, E., & Reyneke, M. (2013). To share or not to share: The role of content and emotions in viral marketing. Journal of Public Affairs, 13(2), 160-171.

Campbell, C., Ferraro, C., & Sands, S. (2014). Segmenting consumer reactions to social network marketing. European Journal of Marketing, 48(3/4), 432-452.

Cannon, W. (1932). The wisdom of the body. New York: Norton cited in: Taylor, S. (1991). Asymmetrical effects of positive and negative events: The mobilization-minimization hypothesis. Psychological Bulletin, 110(1), 67-85.

Change the size or resolution of a picture. (n.d.). In Microsoft. Retrieved from

http://windows.microsoft.com/en-hk/windows-vista/change-the-size-or-resolution-of-a-picture

Chevalier, J. A., & Mayzlin, D. (2006). The effect of word of mouth on sales: Online book reviews. Journal of marketing research, 43(3), 345-354.

Clemons, E. K., Barnett, S., & Appadurai, A. (2007, August). The future of advertising and the value of social network websites: some preliminary examinations. In Proceedings of the ninth international conference on Electronic commerce (pp. 267-276). ACM. De Vries, L., Gensler, S., & Leeflang, P. S. (2012). Popularity of brand posts on brand fan

pages: an investigation of the effects of social media marketing. Journal of Interactive

Marketing, 26(2), 83-91.

Ellison, N. (2007). Social network sites: Definition, history, and scholarships. Journal of Computer-Mediated Communication, 13, 210-230.

Fazio, R. H. (1986). How do attitudes guide behaviour. Handbook of motivation and cognition: Foundations of social behavior, 1, 204-243.

Guadagno, R. E., Rempala, D. M., Murphy, S., & Okdie, B. M. (2013). What makes a video go viral? An analysis of emotional contagion and Internet memes. Computers in Human Behavior, 29(6), 2312-2319.

Gunawan, D.D., & Huarng, K.-H., Viral effects of social network and media on consumers’ purchase intention, Journal of Business Research (2015)

Hansen, L. K., Arvidsson, A., Nielsen, F. Å., Colleoni, E., & Etter, M. (2011). Good friends, bad news-affect and virality on Twitter. In Future information technology (pp. 34-43). Springer Berlin Heidelberg.

Heilman, K. H. (1997). The neurobiology of emotional experience. The neuropsychiatry of limbic and subcortical disorders, 133-142.

Hennig-Thurau, T., Gwinner, K. P., Walsh, G., & Gremler, D. D. (2004). Electronic word-of-mouth via consumer-opinion platforms: what motivates consumers to articulate themselves on the internet?. Journal of interactive marketing, 18(1), 38-52.

Henning-Thurau, T., Gwinner, K., Walsh, G., & Gremler, D. D. (2004). Electronic word-of-mouth via consumer-opinion platforms: What motivates consumers to articulate themselves on the Internet? Journal of Interactive Marketing, 18 (1): 38–52.

Herr, P. M., Kardes, F. R., & Kim, J. (1991). Effects of word-of-mouth and product-attribute information on persuasion: An accessibility-diagnosticity perspective. Journal of consumer research, 454-462.

Ho, J. Y., & Dempsey, M. (2010). Viral marketing: Motivations to forward online content. Journal of Business Research, 63(9), 1000-1006.

Hofstede, G. (1980). Culture’s consequences: International differences in work-related values. Beverly Hills, CA: Sage.

Kabadayi, S., & Price, K. (2014). Consumer–brand engagement on Facebook: liking and commenting behaviors. Journal of Research in Interactive Marketing, 8(3), 203-223. Kaplan, A. M., & Haenlein, M. (2010). Users of the world, unite! The challenges and

opportunities of Social Media. Business horizons, 53(1), 59-68.

Katz, E., & Lazarsfeld, P. F. (1955). Personal Influence, The part played by people in the flow of mass communications. Transaction Publishers.

Kelly, L., Kerr, G., & Drennan, J. (2010). Avoidance of advertising in social networking sites: The teenage perspective. Journal of Interactive Advertising, 10(2), 16-27.

Lane, R., Miller, A. N., Brown, C., & Vilar, N. (2013). An Examination of the Narrative Persuasion with Epilogue through the Lens of the Elaboration Likelihood Model, Communication Quarterly, 61:4, 431-445.

Moran, G. & Muzellec, L. (2014). eWOM credibility on social networking sites: A framework. Journal of Marketing Communications, 1-13.

Mulligan, K., & Scherer, K. R. (2012). Toward a working definition of emotion. Emotion Review, 4(4), 345-357.

Rimé, B., Corsini, S., & Herbette, G. (2002). Emotion, verbal expression, and the social sharing of emotion. The verbal communication of emotions: Interdisciplinary perspectives, 185-208.

Soares, A. N., Pinho, J. C., Nobre, H. (2012). From social to marketing interactions: The role of social networks. Journal of Transitional Management, 17(1), 45-62.

Taylor, D. G., Strutton, D., & Thompson, K. (2012). Self-enhancement as a motivation for sharing online advertising. Journal of Interactive Advertising, 12(2), 13-28.

Taylor, S. E. (1991). Asymmetrical effects of positive and negative events: the mobilization-minimization hypothesis. Psychological bulletin, 110(1), 67.

Trusov, M., Bucklin, R. E., & Pauwels, K. (2009). Effects of word-of-mouth versus traditional marketing: findings from an internet social networking site. Journal of marketing, 73(5), 90-102.

Watson D., & Tellegen A. 1985. Toward a consensual structure of mood. Psychological Bulletin, 98(2), 219–235.

Wu, L. L., Wang, Y. T., & Hsu, A. C. (2014). EWOM EFFECTS ON FACEBOOK.

Wu, P.F. (2013). In search of negativity bias: An empirical study of perceived helpfulness of online reviews. Psychology and Marketing, 30(11), 971-984.

Wu, S., Tan, C., Kleinberg, J. M., & Macy, M. W. (2011, May). Does Bad News Go Away Faster?. In ICWSM.

Xiao, C., Zhou, F., and Wu, Yu. (2013). Predicting audience gender in online content-sharing social networks. Journal of the American Society for Information Science and

Technology, 64(6), 1284-1297.

Appendix