• No results found

Emotional spillover effect in user-generated and firm-generated pictorial content on social media

N/A
N/A
Protected

Academic year: 2021

Share "Emotional spillover effect in user-generated and firm-generated pictorial content on social media"

Copied!
94
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Emotional spillover effect in user-generated and

firm-generated pictorial content on social media

Author: Caroline Michèle Muller

Student ID: 11087412

University of Amsterdam, Faculty of Economics and Business

MSc. Business Administration - Marketing

Date: 24

th

of June 2016

(2)

Statement of originality

This document is written by Caroline Michèle Muller 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)

Table of Contents

Index of Figures ... 5   Index of Tables ... 5   Abstract ... 6   1.   Introduction ... 7   2.   Literature Review ... 12   2.1.   Word-­‐of-­‐Mouth  ...  12   2.2.   Electronic  Word-­‐of-­‐Mouth  ...  13  

2.3.   Visual  cues  in  Electronic  Word-­‐of-­‐Mouth  ...  16  

2.4.   Ad-­‐Evoked  Emotions  ...  18   3.   Conceptual Framework ... 22   3.1.   Conceptual  Model  ...  22   3.2.   Hypotheses  ...  23   4.   Methodology ... 28   4.1.   Research  design  ...  29  

4.1.1.   Selection  of  product  category  and  brands  ...  30  

4.1.2.   Visual  content  collection  and  SentiBank  Analysis  ...  31  

4.1.3.   Measures  ...  33  

4.2.   Data  collection  ...  37  

4.2.1.   Pilot  study  ...  37  

4.2.2.   Main  study  ...  39  

4.2.3.   Data  cleaning  and  Reliability  Analysis  ...  40  

4.2.4.   Manipulation  checks  ...  43  

(4)

5.1.   Correlations  ...  45  

5.2.   Stepwise  Regression  Model  ...  47  

5.3.   Hypotheses  Testing  ...  49  

5.3.1.   Emotional  spillover  effect  ...  49  

5.3.2.   Mediating  effect  ...  50  

5.3.3.   Inter-­‐Correlation  and  effect  of  valence  of  emotional  response  ...  51  

5.3.4.   Moderating  effect  ...  53   6.   Discussion ... 55   6.1.   Key  findings  ...  55   6.2.   Managerial  implications  ...  59   6.3.   Limitations  ...  61   6.4.   Further  Research  ...  64   7.   Conclusion ... 66   References ... 68   Appendix A ... 74   Appendix B ... 79                    

(5)

Index of Figures

Figure 1: Visual Model...22

Index of Tables

Table 1: Factorial Design...29

Table 2: Demographic characteristics of respondents...40

Table 3: Manipulation Checks...45

Table 4: Means, Standard Deviations and Correlations (1)...47

Table 5: Stepwise Regression Model...48

Table 6: Means, Standard Deviations and Correlations (2)...49

Table 7: Mediation (EoP and Att mediated by ER) ...51

Table 8: Inter-Correlations...52

Table 9: Regression...52

Table 10: Moderation (EoP and ER moderated by source) ...53

Table 11: Moderation (ER and Att moderated by source) ...54

Table 12: Moderation (ER and Att moderated by involvement) ...55              

(6)

Abstract

 

The success of social media channels has revolutionized the relationship between brands and consumers. One main factor is the communication between and in-between both parties and with the recent emergence of photo-sharing platforms, visual content has gained importance in this context. Firms and users create brand-related content on social media platforms that is able to modify consumers’ attitude towards the brand. Visual electronic Word-of-Mouth (eWOM) on social media platforms and consumers’ emotional responses to such contents need further consideration.

The purpose of this research is to gain valuable insight into the emotional spillover effect of brand-related pictures on social media, to study differences between firm- and user-generated content and to examine effects on attitude towards the brand. This study undertakes a controlled online consumer experimental survey with 181 international respondents. In this specific experimental setting, Instagram is the selected platform and controversial fashion brands are the chosen target for the visual content. The study verifies the existence of a spillover effect of emotions on the picture on the emotional response. Emotional response acts as full mediator between emotions on the picture and attitude towards the brand. Positive and negative responses have significantly different effects on attitude towards the brand; in fact the relationship is stronger for positive emotional elements than for negative ones. Furthermore, the moderating effect of the source of the content and consumer involvement is proven insignificant. It is suggested that recognizing the source involves further cognitive resources and is not relevant in the context of affective responses. Further research is required to determine specific effects of visual eWOM, their antecedents and consequences. Managers should understand the relevance of visual elements in eWOM and consider emotions while developing online marketing strategies.

(7)

1. Introduction

Social media has become an increasingly important source for marketers to communicate with customers and track their online behaviour (Kozinets, Valck, Wojnicki, & Wilner, 2010; Kumar, Bezawada, Rishika, Janakiraman, & Kannan, 2016). In fact, expenditures assigned to social media advertising worldwide have reached $23 billion in 2015 and will presumably pass $33 billion in 2017 (eMarketer, 2015). In addition to paid online advertising and firm-generated content, positive consumer-firm-generated content can also be regarded as advertising for the firm. The effectiveness of the so-called electronic word-of-mouth (eWOM) has been proven in many studies (Babić, Sotgiu, de Valck, & Bijmolt, 2016; King, Racherla, & Bush, 2014; Ya You, Vadakkepatt, & Joshi, 2015) and its cost-free nature highlights its importance even more.

Word-of-mouth (WOM) has great power to shape customer preferences and is supportive in customer acquisition, especially o social media platforms (Godes & Mayzlin, 2004; Trusov et al., 2009). Managers should actively follow online consumer behaviour because of the positive relationship between eWOM and sales; this inside knowledge should then be incorporated in long-term strategic decisions (Babić et al., 2016). Several aspects of eWOM have already been studied in detail, including the use of different platforms, product categories or metric factors, like volume and valence (Babić et al., 2016). In general, it is focused upon textual eWOM, meaning that other elements, especially visual eWOM, are missing in current literature. In fact, current eWOM studies have not specifically dealt with visual eWOM, including online postings of pictures and videos (Babić et al., 2016; King et al., 2014).

Indeed a strong emphasis has recently been placed on online visual content, especially with the emergence of new social media platforms that are based upon photo-sharing, for instance

(8)

Instagram (Bakhshi, Shamma, & Gilbert, 2014; Thomas, 2012). Instagram has known a huge success and a fast-paced growth since its launch in October 2010. The number of users has grown from 1 million in 2010 to 400 million in 2015 and over 40 billion pictures have already been shared. Instagram observes a daily number of about 3.6 billion likes and a daily average upload of 80 million pictures (Instagram, 2016a). The concept of this social media platform is based on giving people the opportunity to directly share pictures and memories with friends. Filters can optionally optimize snapshots and subsequently be shared on different platforms (Instagram, 2016a). Instagram has created a community where people mainly communicate through visual content. Distribution and communication to a large number of people is made easier than through textual content (Bakhshi et al., 2014).

In fact, firms and consumers use Instagram as platform to create brand-related content and to communicate with each other. It needs to be considered that visuals in eWOM could have a different impact on consumers than mere text. Visual cues produce different level of attention, each having a different outcome on product and brand evaluations (Pieters & Wedel, 2004; Rosbergen, Pieters, & Wedel, 1997). In advertising, it is known that compared to textual elements, pictorial elements are superior in catching attention (Pieters & Wedel, 2004). Visual content can also influence recall and consumer involvement with the brand (Rosbergen et al., 1997).

As eWOM is dependent on consumers’ experience and evaluation of the brand, content can be positive, neutral or negative (Liu, 2006). The valence of eWOM is supposed to influence sales, the precise impact is however debated among researchers (Babić et al., 2016; Chevalier & Mayzlin, 2006; Zhang, Craciun, & Shin, 2010). Valence might also be a deciding factor in visual eWOM as it is supposed to be closely related to emotional response. A reaction to visual content is more intuitive and can elicit immediate emotional response (Mitchell, 1986).

(9)

Emotional response towards a brand or product can essentially modify brand-related judgements and attitude towards the brand (Edell & Burke, 1987, 1989; Howard & Gengler, 2001). As emotions on pictures or advertisements can have an effect on product and brand evaluations, the spillover effect of emotions is essential in this matter (e.g. Hasford et al. 2015; Schwarz & Clore 2003; Holbrook & Batra 1987). In addition, it is possible that the spillover effect differs for positive versus negative emotions. Emotional spillover has not been researched based on valence of emotions and distinctive effects can thus be expected (Howard & Gengler, 2001). Overall, visual eWOM might have distinctive impacts on consumer behaviour, evaluation of eWOM as well as the brand.

The area of visual cues in online word-of-mouth communication is rather unexplored and has not been researched yet, especially in relation to emotional spillover effects on social media channels. Building on the above-mentioned arguments, the objective of this study is to research how positive and negative emotions represented in brand-related pictorial content on social media affect consumers’ emotional response and the attitude towards the brand. Brand-related content can be created by users and by the firm; a distinction needs to be made. Firm-generated content in social media can have fundamental and positive outcomes on profitability and spending of consumers (Kumar et al., 2016). For that reason, it will be observed whether and how the effect differs between user-generated and firm-generated content. Different emotions represented on the pictures and the corresponding contagion effect will be taken into account to evaluate this effect. It will be interesting to observe whether and in which conditions the emotions transmitted in the pictures will affect the emotional response of the receiver and thus have an impact on attitude towards the brand. As a consequence, the effect on brand evaluations and the moderating role of previous involvement can also be identified.

(10)

In order to encounter this lack in current literature, specific research questions were constructed and covered in the course of this study. The primary research question that will be addressed in this paper is “If and how does the emotional spillover effect of brand-related

pictures on social media differ for UGC versus FGC?”. Secondary questions in this study

will address “How does the emotional response influence the attitude towards the brand?” and “Whether and how does the level of consumer involvement influence this relationship?”.

The results of this study can have key managerial implications. Electronic interpersonal communication offers opportunities and threats for companies. Marketers need to constantly update their knowledge about online consumer behaviour to design effective digital marketing strategies. As visual content in social media becomes increasingly important, marketers should be aware of its effects on customers and their emotional response to the brand. Knowing which type of content considerably increases the engagement of the online community can be crucial (Bakhshi et al., 2014). Recognizing whether firm-created or user-created content is more fruitful to deliver emotions can help marketers to develop an improved marketing strategy. Some companies tend to build their strategy on the theory that a small number of individuals is capable of influencing a considerable part of the population (Katz & Lazarsfeld, 1955). In the process, firms intend to sway these influencers as they are supposed to act as mediators between the firm and the mass of (potential) consumers. The significance of these influencers and their ability to cause cascades have however been criticised (Watts & Dodds, 2007). For that reason, new insights will support firms in choosing an optimized online marketing strategy. In addition, understanding which type of emotions can easily be transferred while positively affecting users’ evaluation about brands can be incorporated in the firm’s social media strategy. Companies can figure out and foster the most influential content.

(11)

This paper can provide new insights for academic literature and can give incentives for further research in this area. Babić et al. (2016) receently published a meta-analysis recently, outlining that most studies deal with textual eWOM while other dimensions of eWOM have been neglected. Content as well as consumer responses in this context have not been fully explored yet and need further research (Babić et al., 2016). Visual eWOM has barely been researched but its importance has been outlined in recent literature analysis. There is still no knowledge about how visual eWOM is processed, how consumers interpret this content and how its credibility is evaluated (King et al., 2014). In addition, the importance of emotional contagion in advertising and marketing has been proven but the relation to WOM and eWOM is rare. In fact, eWOM is shared more frequently if the content elicits high-arousal emotions, independent of their valence, and thus increases its effectiveness (Berger & Milkman, 2012). WOM and eWOM are not restricted to the sharing of information, but sharing and transmitting emotions can occur as well and give explanation for the effectiveness and results of WOM (Howard & Gengler, 2001).

First of all, this paper outlines the literature review essential for this topic. This section covers definitions and explanations about WOM, eWOM, visual cues in eWOM and ad-evoked emotions. The following part consists of the conceptual framework where the visual framework, the variables as well as the hypotheses will be introduced. In the methodology part, research design and the data collection are described. Research design includes the research method, the selected brands and the measures used in this paper. Data collection covers data cleaning, results of the reliability analysis and the manipulation checks. Hypotheses are tested and the results of correlation and regression analysis are presented. Discussion including key findings, managerial implications, limitations and further research are outlined as final part of the paper followed by a conclusion.

(12)

2. Literature Review

2.1. Word-of-Mouth

 

“Powerful networks of interpersonal relations existing within the consumer market (Brooks, 1957, p. 154)” are a crucial source for collecting information about products (Brooks, 1957). Interpersonal communication, referred to as Word-of-mouth (WOM), is a powerful tool for consumers to receive and exchange information about products and brands (Kozinets et al., 2010). Studies confirm that interpersonal communication has a significant impact on modifying opinion and behaviour. Some individuals can act as mediator between the media and a large part of the society (Katz & Lazarsfeld, 1955). So-called opinion leaders, consumers with a special interest and knowledge in a certain field, are powerful in spreading product experiences and can induce attitudinal and behavioural changes (Brooks, 1957). Purchasing decisions can also be altered and can thus have an effect on sales (Chevalier & Mayzlin, 2006). The pivotal role of opinion leaders has been challenged though; it seems to be more fruitful to target individuals that are influenced more easily. These susceptible individuals will consequently impact each other and possibly trigger cascades (Watts & Dodds, 2007). Overall, emphasis is put on interpersonal relations and communication between individuals, irrespective if they are influencers or not. WOM is diffused throughout various communities and can increase awareness as well as shape preferences. Therefore it seems beneficial for firms to stimulate well-diffused WOM online and offline (Godes & Mayzlin, 2004).

The sender of WOM can have different motivational factors, like product involvement, self-involvement, concern for others and dissonance reduction (Angelis, Bonezzi, Peluso, Rucker, & Costabile, 2012; Engel, Kegerreis, & Blackwell, 1969).

Besides the existence of traditional WOM, which is about interacting face-to-face in different social contexts and groups, a new stream of WOM has emerged where customers trust

(13)

written messages from unknown individuals in an online environment (J. Brown, Broderick, & Lee, 2007). Members of online communities show the same pattern of social influence between members than those in offline communities (Trusov, Bodapati, & Bucklin, 2010). 2.2. Electronic Word-of-Mouth

 

Today, more than 3 billion people worldwide use the Internet, around 7 billion devices have online connection and nearly every second household has Internet access. In the last 15 years, Internet diffusion marked a remarkable increase from 6.5% to 43% underlining its inevitable importance in the contemporary environment (International Telecommunication Union, 2015). The impressive growth of the Internet has substantially modified relationships between organizations and customers as well as among customers. The significant expansion, importance and prevalence of electronic Word-of-mouth communication has aroused interest among marketers and researchers (Babić et al., 2016; J. Brown et al., 2007; King et al., 2014; Ya You et al., 2015).

Online WOM builds on the same concept than offline WOM; the main difference is the informational spread via Internet. This communication can take different forms; e.g. tweets, posts, likes, blog posts, hash tags, reviews, pictures and videos. Consumers’ decision-making process has been altered significantly due to increasing eWOM (Babić et al., 2016). Online information searching, online shopping and social media platforms have become increasingly popular and place emphasis on eWOM’s relevance (King et al., 2014). This advanced online landscape impacts “brand building, customer acquisition and retention, product development, and quality assurance (Dellarocas, 2003, p. 1407)”. Main reasons for customers to actively look for eWOM is to reduce uncertainty and risk as well as getting a wide range of information before purchasing a product (Babić et al., 2016).

(14)

Researchers agree on the fact that eWOM impacts sales because purchasing decisions are closely related to sales, the precise impact is however still debated (Babić et al., 2016). The valence of eWOM can be positive, negative or neutral (Liu, 2006).

On the one hand, some studies emphasise a jeopardizing effect on sales by negative eWOM (Chevalier & Mayzlin, 2006). On the other hand, positive eWOM can be more effectual than negative eWOM and negative information might not directly be connected to a reduction in sales. Negative effects on sales can only be spotted in later product lifecycle stages and in cases with low financial risks (Babić et al., 2016; Zhang et al., 2010).

Differences can be observed in product, industry and platform characteristics as well as the metric factors, namely volume and valence (Babić et al., 2016; Ya You et al., 2015). The importance and capability of eWOM to modify consumers’ opinion is undeniable because it seems to have a greater short-term elasticity than other marketing-mix tools (Ya You et al., 2015) .

Interaction between consumers is more pronounced on social media than on other platforms and therefore motivates to relationship building and identity revelation (Ya You et al., 2015). Social media or social networking sites are important tools for sharing user-generated content (UGC) and consumer communication on the Internet; it enables marketers to study WOM easier and more accurately than ever before (Trusov et al., 2009). While comparing eWOM on social networking sites (SNS) to traditional marketing communication, Trusov et al. (2009) revealed that the former has a greater impact on acquiring new customers and that the effects are long-term. They even categorize WOM as one of the most effective marketing communication strategies (Trusov et al., 2009). The efficacy of social media platforms depends however on the similarity between the sender and the receiver of eWOM. In general, the effect of eWOM is stronger if the similarity between both parties is more pronounced.

(15)

This means that managers should encourage consumers to release detailed information about themselves (Babić et al., 2016). Identifying highly influential online users can be a supporting factor in the targeting process and can turn out to be rather cost-effective. Opinion leaders have the ability to stimulate the usage level of others by increasing their own. This only concerns individuals who are connected to the influencer on the social media platform (Trusov et al., 2010).

The above-mentioned effects of eWOM refer to textual content, visual eWOM might alter, strengthen or weaken the effects or have additional outcomes. These implications will be further developed in the next section.

An increasing number of companies use consumer-to-consumer communication as a marketing technique. It became much easier to actually follow and monitor what consumers communicate about companies and thus react or try to influence accordingly (Kozinets et al., 2010). Companies profit from this low-cost method to contact customers at a large-scale whereas customers benefit from the opportunity to express their attitudes, conceptions and response throughout the online network (Dellarocas, 2003).

In addition to user-generated content on social media, firms have their own social media profiles and produce firm-generated content (FGC). FGC is defined as “firm-initiated marketing communication in its official social media pages (Kumar et al., 2016, p. 7)”. It is a rather unexplored field in marketing research mainly because the additional value of such unknown advertising is difficult to measure. Nonetheless, the existence of differences in reactions to user- and firm-generated ads is acknowledged (Pehlivan, Sarican, & Berthon, 2011). Kumar et al. (2016) analysed the effect of FGC on customer spending, profitability and cross-selling activity. A positive relationship with customer profitability has been discovered, which means that FGC can be used to build worthwhile relationships and create

(16)

one-to-one interactions with consumers. These enhanced relationships may even have positive impacts on sales. Consumers can, besides only absorbing, share, comment and “like” the content and thus create favourable brand evaluations. It though seems to be more than solely an advertising instrument. It actually functions alongside television and e-mail marketing and encourages cross-buying behaviour. They revealed proof that FGC has a positive effect especially on customers who are experienced social media users and on those who already have a long-term relationship with the firm. The effect however depends on several factors like for instance the valence of the message, the reaction of the consumer and their initial opinion towards social media channels (Kumar et al., 2016).

2.3. Visual cues in Electronic Word-of-Mouth

Visual illustrations are superior to words when it comes to representing reality. Social media platforms with the primary goal of sharing visual content namely Instagram, Tumblr and Pinterest have experienced a huge success over the last years (Thomas, 2012). The upcoming trend of posting and sharing visual content on social media has raised questions regarding the relevance of brand-related visual eWOM. The emergence of this specific field is recent and thus poorly researched (King et al., 2014).

Advertising literature can be used to elaborate on the effects of visuals. In advertising, visual elements are more frequently in focus than informational elements and emphasis is put on the provoked emotional response (Mitchell, 1986). Rosbergen et al. (1997) outline that visual cues have different effects on consumers depending on their attention or motivation to process the information. In advertising, visual cues are composed of various physical characteristics of a specific advertisement. Rosbergen et al. (1997) were able to define customer segments with similar visual attention, which have different outcomes on product involvement, brand attitude and advertising recall. The role of headline, picture, packshot and

(17)

their respective position in the advertisement were researched in this context (Rosbergen et al., 1997). Images can transmit more than just the actually illustrated content. Peracchio et al. (2005) mention that stylistic features like the angle of the camera and the position of items can impact descriptive concepts that have an effect on perception. The effect arises if observers are sufficiently able to process the ad and if a suitable descriptive concept is accessible (Peracchio & Meyers‐Levy, 2005). Pieters et al. (2004) explain that in print advertisements a distinction is made between brand-related, pictorial and textual elements. Each of these components can have particular effects on catching consumers’ attention and pictures seem to be most beneficial. Higher attention in general means that the ad and its elements are observed more carefully and are therefore easily remembered (Pieters & Wedel, 2004).

The impact of visual content cannot be overlooked as it becomes increasingly important on social media platforms. Pictures contain additional information that goes beyond than the accompanying textual description. In order to fully understand posted visual content on social media, an image-based analysis of emotions, sentiments and affect is necessary (Borth, Chen, Ji, & Chang, 2013). Visual content with strong sentiments can encourage the depicted belief and can have a heightened effect on the public (T. Chen, Borth, Darrell, & Chang, 2014). In fact, consumers can share textual as well as pictorial content about brands. Pictorial content on Instagram contains other information besides the visual element, as it is often combined with tags, likes, hashtags and comments. Visual eWOM can therefore be put on a level with pictures in advertisements and can then be acknowledged as a type of user-generated ad. This definition is important when relating visual content in ads to emotions in the next section.

(18)

2.4. Ad-Evoked Emotions  

Understanding consumer’s cognitive response to advertising can give deep insights into attitudes towards the ad and the brand. However, these attitudes are also modified by consumers’ direct affective response. Feelings and emotions are necessary to interpret the impact and effectiveness of an ad (Edell & Burke, 1987).

It is very difficult to compose a clear definition for emotions. Emotions can be referred to as “complex functional wholes including appraisals or appreciations, patterned physiological processes, action tendencies, subjective feelings, expressions, and instrumental behaviours (Fischer, Shaver, & Carnochan, 1990, p. 85)”. Hatfield et al. (1994) state that emotions can either be positive or negative and can influence behaviour as well as the nervous and internal body system. People express their emotions using their face, voice and body language. All these elements can actually occur simultaneously or subsequently (Hatfield, Cacioppo, & Rapson, 1994).

Berger (2014) affirms that one general function for the sender of WOM is emotion-management. He explains that consumers try to gain support, to reduce negative feelings, to understand their feelings, to lessen dissonance, to take revenge or to relive positive experience. Many users might however avoid sharing negative content and emotions because they don’t want to appear as negative people. Positive content is shared with the goal to create an even more favourable feeling and to relive the positive emotions provoked by the brand or product (Berger, 2014).

The valence of emotions and the level of arousal can shape the diffusion of online content (Berger & Milkman, 2012; Berger, 2014). Positive emotions create more virality in the online environment than negative emotions. Consumers are more motivated to share positive

(19)

content as it is connected with a favourable self-representation and identity (Berger & Milkman, 2012; Berger, 2014). However, even if emotions have the same valence, they can have different effects on virality depending on different activation levels. Content triggering high-arousal emotions (e.g. anger, anxiety, awe) is more likely to become viral than content triggering low-arousal emotions (e.g. sadness) (Berger & Milkman, 2012).

Emotions can be classified on different levels; the first level distinguishes between negative and positive emotions. Concerning the second level, researchers do not agree on consistent definition (S. P. Brown & Stayman, 1992; Edell & Burke, 1989; Fischer et al., 1990; Holbrook & Batra, 1987; Pham, Geuens, & De Pelsmacker, 2013). Holbrook et al. (1987) classify emotional responses into three dimensions: pleasure, arousal and domination. Edell and Burke (1989) define between upbeat, warm and negative emotions. Upbeat and warm feelings result in a positive attitude towards the brand and the ad, whereas negative feelings evoke negative judgements (Edell & Burke, 1989). Positive affect can be grouped into contentment, happiness, love and pride whereas negative affect into anger, fear, sadness and shame (Laros & Steenkamp, 2005). Pham et al. (2013) combine different studies in order to organize positive feelings into warmth, excitement and happiness/cheerfulness categories. By relating emotions, cognition and action, Plutchik (2001) discovered that these three concepts take feedback loops. Certain situations provoke emotions, which drive people to take action. In his model, he outlines the relationship between different emotions in his model and defines eight primary emotions. These emotions are fear, anger, joy, sadness, acceptance, disgust, expectation and surprise. They have different intensities, which means that every one of them has one corresponding emotion with lower intensity and one with higher intensity (Plutchik, 2001).

(20)

Feelings induced by advertisements are able to influence the evaluation of the ad and of the brand (Edell & Burke, 1987, 1989; Holbrook & Batra, 1987). Emotions can be seen as mediators towards the response to advertising and towards the response to brand attitude (Holbrook & Batra, 1987). The mediating role of emotions in advertising can be complemented by the concept of emotional contagion/emotional spillover. This concept explains that the emotional stimuli “arise from one individual, act upon (…) one or more other individuals, and yield corresponding or complementary emotions (…) in these individuals (Hatfield et al., 1994, p. 5)”. This process can simply occur as a natural response without control or awareness, which results in imitating expressions and behaviour and thus feeling these expressed emotions (Hatfield et al., 1994).

Due to the spillover effect, emotions of the sender can mingle with the emotions of the receiver and can also alter product attitudes. A happy sender can significantly and positively influence the receiver’s emotion and product evaluation (Howard & Gengler, 2001). This effect is emphasised further if the receiver is actually able to see the smile of the sender. Emotional contagion is frequently connected to the ability of mimicking facial expressions (Howard & Gengler, 2001). In textual eWOM, emotional spillover cannot occur as such but it may experience increased relevance in the context of visual eWOM and photo-sharing social media platforms.

Consumers tend to use heuristics to evaluate products. Heuristics are mental shortcuts that simplify decision-making and problem solving (Petty & Cacioppo, 1983). Motivation and ability to process information influence how consumers change attitudes. Mental shortcuts drive individuals to use cues to make decisions easily and change their attitude accordingly (Petty & Cacioppo, 1983). People tend to use feelings as source of information, which causes

(21)

them to assign their emotions to specific targets. The affect-as-information theory has become prevalent in business literature as it can be related to consumer judgments (Schwarz & Clore, 2003). In fact, emotions are often used as stimuli to influence consumers’ evaluation of products and brands in communication and advertising (Hasford et al., 2015). Hasford et al. (2015) prove that emotional contagion can even have an effect on unrelated products. The valence of emotions expressed by a spokespeople in an advertisement can have a significant effect on brand evaluations. Moderators of the attitude change monitor feelings and emotional ability (Hasford et al., 2015).

Edell and Burke (1987, 1989) state that positive and negative feelings can arise simultaneously, can both forecast the effectiveness of the advertisement and influence the attitude formation towards the ad and the brand. They both provide individual contributions (Edell & Burke, 1987, 1989). Positive ad evoked-feelings have a more distinctive effect on hedonic products than on utilitarian products (Pham et al., 2013). Mitchell (1986) affirms that the valence of affect-laden visual content in ads can have an effect on attitude towards the advertisement and brand attitude. He declares that positive pictures create favourable brand attitudes whereas negative pictures induce unfavourable brand attitudes and indicates that this same phenomenon holds for attitudes towards the advertisement, in fact the effect is even more pronounced. The valence of visual content can thus be equated with an emotional stimulus and has the power to moderate attitude towards the ad. Attitude formation is however influenced by the whole ad and not only by pictures (Mitchell, 1986). Some minor product-unrelated information can have a worthwhile effect on consumers, consequently beliefs about the brand can be deducted from visual elements (Mitchell & Olson, 1981; Mitchell, 1986).

(22)

Plutchik’s (2001) classification of emotions will be used in this study in order to differentiate between different emotions. On the first level, the classification will be used to differentiate between positive and negative emotions. Positive emotions will consider joy, acceptance, expectation and surprise as a second layer. Negative emotions will be composed of fear, anger, sadness and disgust.

3. Conceptual Framework

In this section, different variables used in this study and their assumed relationships and effects will be explained by developing the conceptual framework. First, the visual model (Figure 1) is introduced to give an overview of the different variables and will then help to form the different hypotheses in the second part.

Figure 1: Visual Model

3.1. Conceptual Model

The visual model (Figure 1) for the conceptual framework mixes elements from Word-of-mouth, the ad-evoked emotions and the emotional contagion literature. For the WOM element, the sender, the message and the receiver are essential. In this study, the sender has two different values, namely firm or consumer, and the receiver only one, namely consumer. The emotions sent by the source are depicted in the visual content posted on Instagram. These emotions can in total either be positive or negative and will provoke an emotional

(23)

response from the receiver, which can again either be positive, negative or neutral. The emotions sent and the emotions received will determine the spillover effect of emotions. If they are in accordance, a spillover effect can be proven. The source of the content can modify the association between emotions on the picture and/or emotional response and between the latter and attitude towards the brand. Additionally, the valence of emotional response elements might strengthen or weaken the relationship between emotional response and attitude towards the brand. Previous level of consumer involvement with the studied brand might also affect and moderate this relationship.

3.2. Hypotheses  

Hypothesis 1

The definition of emotional contagion/spillover effect given above gives insight into how emotional stimuli can influence emotional response. An emotional stimulus can either be positive or negative and can induce people to react accordingly. This synchronous behaviour occurs accidentally and uncontrollably (Hatfield et al., 1994). Especially positive emotions elicited by the sender have a powerful impact on receivers’ emotional response. A positive emotional stimulus will influence the emotional response positively as well (Howard & Gengler, 2001). Hatfield et al. (1994) do not differentiate between the valences of emotions. Negative celebrities in advertising can stimulate negative emotional response, allowing the occurrence of an emotional contagion (Hasford et al., 2015). Negative emotional stimuli, like sad faces, generate an especially synchronous behaviour as receivers start to have negative feelings (Small & Verrochi, 2009).

The definition of emotional contagion and the previously summarized information about positive and negative emotions allow the assumption that the emotional response will correspond with the previous valence of the emotional stimuli.

(24)

H1: There is existence of an emotional spillover effect, meaning that there is a positive relationship between positive emotions on the picture and positive emotional response and between negative emotions on the picture and negative emotional response.

Hypothesis 2

The receiver’s emotional response towards the user or firm-generated visual content can illustrate an important step towards the evaluation of the brand. Edell and Burke (1987, 1989) confirm that emotions play a particular role in forming or influencing attitude towards a brand. There exists a direct and an indirect effect of feelings on attitude towards the ad and the brand (Edell & Burke, 1987, 1989). Additionally, the mediating role of emotions in advertising has been proven. The emotional response mediates the link between the content attitude and the attitude towards the ad and the brand (Holbrook & Batra, 1987).

As visual eWOM was previously defined as a sort of user-generated ad and FGC is equally considered as advertising, the role of emotional response will hold in this context. The emotional response to the visual online content will therefore have a mediating effect on the attitude towards the brand.

H2: The positive relationship between the emotions in the visual content and the attitude towards the brand is mediated by the emotional response.

Hypothesis 3

The visual element can provoke positive and negative emotional responses. In general, negative WOM is known to have a more powerful impact on product evaluations than positive WOM (Laczniak, DeCarlo, & Ramaswami, 2001; Mizerski, 1982). Unfavourable product evaluations have a significantly stronger effect on affect towards the product, its performance and credibility than favourable evaluations (Mizerski, 1982).

(25)

Moreover, it is known that emotions have an effect on consumer’s brand attitude (Holbrook & Batra, 1987). It is known that consumers use emotions as a basis for decision-making; they evaluate their emotional state and judge a product accordingly while forming their brand attitude (Schwarz & Clore, 2003). Attitude towards the brand is formed according to the type of emotion connected to the specific brand. Edell and Burke (1987) give further details about the various effects of emotional valence. Positive feelings and a previous positive attitude influence attitude toward the ad favourably. However, negative feelings have a higher effect on the attitude towards the ad than positive feelings. Ad-evoked negative feelings are especially important for the evaluation of the brand, even if a previous attitude did exist. If no previous attitude exists, evoked feelings can be enough to generate certain beliefs (Edell & Burke, 1987). The effect of negative emotions on attitudes is always negative (Edell & Burke, 1989).

By combining the knowledge on the effects of WOM valence and emotional valence, negative elements seem to have a more intense effect on attitudes and brand evaluations.

H3: The positive relationship between emotional response and the attitude towards the brand differs for positive and negative emotional response, so that the effect is stronger for negative emotional response than for positive emotional response.

Hypothesis 4

Referring to the literature, WOM and eWOM are counted among the most effective tools in affecting behaviour and attitudes of consumers. Consumers use social media to communicate their opinion and experience about brands and have the potency to significantly change attitudes and behaviour of other consumer (Brooks, 1957; Kozinets et al., 2010). Since consumption always comes with some sort of affective response to the product and WOM is used to share product or brand-related experiences, it is inevitable that eWOM contains

(26)

emotions (Howard & Gengler, 2001). FGC can also positively influence consumer behaviour and advertisements are strong in provoking emotional reactions of consumers (Edell & Burke, 1987, 1989; Holbrook & Batra, 1987; Kumar et al., 2016).

Building on the fact that one of the main reasons for consumers to share content online is to regulate their emotions, it can be assumed that UGC is filled with emotions. WOM and UGC have a social bonding function where communication between individuals can strengthen ties and intensify the emotional transfer (Berger, 2014). Users are actually willing to trust unknown online users, especially when other trustworthy sources are missing (Berger, 2014; J. Brown et al., 2007). As traditional sources of information, like media and advertising, have lost in credibility, consumers rely on each other even more (Berger, 2014). Due to higher level of credibility, positive online consumer reviews seem to be more impactful than firm-produced communication (Ho-dac, Carson, & Moore, 2013).

Motivated by the strength of interpersonal communication and effectiveness of eWOM, it can be expected that eWOM is not only more credible but also more emotional. Therefore in the context of brand-related content, UGC will be more convincing and will thus more effectively transfer emotions than FGC.

H4: The positive relationship between the emotions in the visual content and the emotional response is moderated by the source of the content, so that the relationship is stronger for UGC than FGC.

Hypothesis 5a

As the emotional spillover effect might differ for the two sender sources, the emotional response to the pictorial content may change their attitude towards the brand. Pehlivan et al. (2011) discovered that consumers seem to respond differently to firm-generated ads as opposed to consumer-generated ads in terms of approached subjects in their comments. Even

(27)

though there is no effect based on the source, some elements differ and therefore evoke different responses for both sources. They claim that UGC is more humorous, especially liked by the crowd, and that responses to FGC are frequently related to the description of the product itself and produce more sentiment-related messages (“love” or “hate”). Receivers of eWOM seem to focus more on the description of the ad itself than on the product (Pehlivan et al., 2011). In addition, it is known that the valence of FGC affects consumer behaviour, like spending and cross-buying (Kumar et al., 2016). The effect of the valence of a message can either be attached to the brand and/or to the sender. Brand evaluations improve if the negative message in the content is related to the sender and declines when related to the brand itself (Laczniak et al., 2001).

Reactions to FGC seem to be have a closer link to the brand or the product than responses to UGC (Pehlivan et al., 2011). In this context, the initiator of the message is the brand in question and emotional response will therefore be attached more easily to the brand. Messages in eWOM and the according emotional response might not be connected to the brand as easily because the sender is different. The strength of the relationship between emotional response and attitude might therefore be moderated by the source of the content. The response to the brand might be stronger for FGC than for UGC because a more important part of emotional response is directed towards the brand for FGC than for UGC.

H5a: The positive relationship between emotional response and the attitude towards the brand is moderated by the source of content, so that the relationship is stronger for FGC than for UGC.

Hypothesis 5b

Pan et al. (2011) indicate that the previous level of involvement with the brand or the product can bias the effectiveness of ad-evoked emotions and its corresponding brand evaluation.

(28)

This effect is closely related to the motivation and ability of consumers to deal with information. Consumers already involved with the brand are less likely to change their attitude and handle information more thoroughly. They are motivated to further process information and therefore more likely to react according to their previous attitude (Pan & Zhang, 2011). Low involvement in a brand or product initiates decision-making based on mental shortcuts, which means that ad-evoked feelings will have a pronounced effect. Brand attitude will change or be formed according to the provoked emotions, as low involved consumers are less motivated to analyse information attentively (S. P. Brown & Stayman, 1992; Petty & Cacioppo, 1983; Pham et al., 2013).

In the context of visual content and emotional spillover, the level of involvement prior to exposure may affect the emotional response, based on the level of motivation and ability to examine the content, and therefore influence attitude towards the brand.

H5b: The positive relationship between emotional response and attitude towards the brand is moderated by the level of involvement with the brand, so that this relationship is stronger for lower values of involvement and weaker for higher values of involvement.

4. Methodology

In this section, the methodology of this paper will be introduced with a short description of the chosen research design and the research process. Afterwards, it will be elaborated on the selection of brands for the study and the usage of the visual sentiment analysis SentiBank. Each dependent and independent variable will be defined clearly, operationalized and the measurement method outlined. To complete this section, the data collection including a pre-test and the actual study with first demographic results are presented. Data cleaning methods as well as the reliability checks of the different constructs is explained, followed by the analysis of the manipulation checks.

(29)

4.1. Research design

As hypotheses will be developed and tested on a representative sample, a quantitative method is highly appropriate for this research. A controlled online consumer experimental survey will be the method used for this study. The experimental survey will consist of a 2x2 factorial design (Table 1), where the valence of emotions (positive or negative) and source of the generated content (firm or user) can be modified. Every participant will only participate in one treatment group, which means that every participant will be either in posFGC, negFGC,

posUGC or negUGC.

Positive emotions Negative emotions Firm-generated (pictorial) content Treatment 1 Treatment 2

User-generated (pictorial) content Treatment 3 Treatment 4 Table 1: Factorial Design

Firstly, some requirements for the sample of respondents need to be fixed. Demographics of U.S. Instagram users show that more than half of the users are between 18-29 years old and about one quarter is between 30-49 years old (Duggan, 2015). The sample of this study will mostly consist of tech-savvy students and young professionals aged between 18 and 29 years old. Secondly, firm-generated and user-generated pictures will be extracted from the social media platform Instagram and will be rated on different emotions using a new visual sentiment analysis approach (Visual Sentiment Ontology, SentiBank). A set of controversial brands will be chosen where firm-generated content can also have negative emotional connotations. For every picture, the emotions will be known from the Sentibank so that they can be assigned to the different treatment groups. In a last step, the experiment will be conducted with the approved sample of participants. An online survey with about 50 participants in each treatment group will be conducted to reach a planned total of about 200 participants. Respondents will be exposed to a number of brand-related pictures (either firm

(30)

or user-generated) followed by the same question after every exposure in order to measure emotional response. Using the same method in every treatment group, the emotional response of respondents to each picture will be measured followed by measurements of attitude towards the brand as well as consumer involvement and brand familiarity.

4.1.1. Selection of product category and brands  

Brands and their advertisements vary in their level of controversy. Controversial topics provoke different, strong and even opposing opinions. Controversial issues are very subjective on interpretation and are often related to morality. Controversy can stimulate interest on the one hand, which leads to more WOM, and can arouse discomfort on the other, which lessens the level of WOM. If the controversial level is moderate, the issue is more likely to create more WOM (Z. Chen & Berger, 2013). Overall, controversial topics and thus controversial advertising are able to create different emotions and strength of emotions. Emotions induced by advertising highly affect brand evaluation if they are based on a hedonic rather than on a utilitarian consumption purpose (Pham et al., 2013).

Sometimes so-called shock tactics are implemented to catch people’s attention by using provocative images or slogans. This technique is often used to attract attention to a specific cause or issue, but also for commercial reasons (Advertising Standards Authority, 2015). Shock or controversial advertising often leads to violation of Advertising Codes or Standards like the ASA (Advertising Standards Authority) or the Stiching Reclame Code (Dutch Advertising Code). Based on this knowledge, the industry chosen for this research is the fashion industry.

The fashion industry can be very controversial and provocative, especially in terms of advertising. Based on the previous findings that consumption reasons are closely related to the product category, apparel is likely to arise from hedonic reasons as it originates from

(31)

(Hirschman & Holbrook, 1982). Fashion and apparel are classified as “high culture products within high popular culture (Hirschman & Holbrook, 1982, p. 95)” and thus originate from hedonic consumption reasons.

Four fashion brands known for their controversial advertising are selected for this study, namely Benetton, SuitSupply, American Apparel and Calvin Klein. The Italian fashion brand Benetton (United Colors of Benetton) is well known for provocative advertisements. Their ads range from kissing politicians respectively religious heads, to HIV positive or anti-racism campaigns (The Guardian, 2011). The Dutch brand SuitSupply violated the Dutch Advertising Code and was accused of sexist advertising and misogyny (Reclame Code Commissie, 2014). The U.S. brand American Apparel has caused quite a stir with their advertisements. The brand has been accused of sexualising children and under-aged women, sexist or even pornographic representations and objectifying women (Advertising Standards Authority, n.d.). Some complaints have also been filed against the global lifestyle brand Calvin Klein based on their underweight models or sexuality in their ads broaching the issue of “sexting” and the hook-up culture (Advertising Standards Authority, n.d.; Fitzpatrick, 2015).

4.1.2. Visual content collection and SentiBank Analysis  

In a first step, the pictorial content of the four selected brands were extracted from the social media platform Instagram. The image collection was based on a number of requirements to distinguish between user and firm-generated content. For firm-generated content, information about the different brand accounts on Instagram was gathered. Some brands were present with multiple accounts in different countries, so to assure a consistency in the image collection, only the official global account for each brand was selected for further analysis.

(32)

Firm-generated content was thus based on the official brand accounts namely, “benetton”, “suitsupply”, “calvinklein” and “americanapparelusa”. A total of 8760 images were collected under which 4756 for American Apparel, 1208 for Benetton, 1484 for Calvin Klein and 1312 for Suit Supply; this represented the full amount of content on Instagram at the time of collection.

For user-generated content, three popular hashtags were defined for each brand, which were used to filter out brand-related postings by consumers. Based on this requirement, a total of 3767 Instagram postings were collected under which 551 for American Apparel, 637 for Benetton, 1548 for Calvin Klein and 1031 for SuitSupply. Overall, the collected database included a total of 12527 images and gives a variety of information. On the one hand, information about the accounts, namely user profiles, names and ID’s are presented and on the other hand information linked to the images, namely the number of likes, tags and comments, the image caption, the filter used, the creation date and the URL to the images are provided.

In a second step, the sentiment analysis for this selected database was run through to define the level of emotions and the valence for each picture. It is important to have a clear idea about the emotions presented on the pictures before choosing which ones to use for the experimental survey. The approach used for the sentiment analysis is fairly new and builds on Plutchik’s Wheel of Emotions as a psychological theory (Plutchik, 2001). It consists of a large-scale Visual Sentiment Ontology (VSO) with a large base of Adjective Noun Pairs (ANP). The latter combines nouns with adjectives to give neutral words a sentiment-based meaning. The SentiBank detector library can then spot different ANP’s on images and can identify the sentiments in the pictorial content in the process (Borth, Chen, et al., 2013; Borth, Ji, Chen, Breuel, & Chang, 2013). To establish a meaningful score, fifty APN’s were

(33)

selected for each picture, and then the mean of the sentiment scores was calculated and used as an indicator for the overall valence of sentiment in the picture. Pictures that were not related to any of the brands, due to a different meaning of the hashtags, as well as content created by commercial users (e.g. web shops), were excluded from the user-generated content selection.

The pictures were then ranked according to their number of likes, as the “Search & Explore” section of Instagram is linked to the number of likes. In this section, users can see varying content posted by different users. Instagram choses the content that is especially interesting for a specific user, based on their Instagram usage, and also choses on the number of likes of the posting (Instagram, 2016b). In addition, virality of online content and high-arousal emotions are closely linked in the way that activated emotions create more virality (Berger & Milkman, 2012). A high number of likes can be linked to virality of the content and can thus be an indicator for the arousal of emotions.

The most liked negative pictures and the most liked positive pictures were selected for each UGC and FGC for each brand. For UGC, some users appeared more than once in the top liked list, which lead to the decision to only take one picture from each user to eliminate effects related to liking or disliking the specific user. Overall, 20 pictures were chosen per treatment, as the treatments are tested for every brand separately, and in total of 80 pictures was selected for further analysis.

4.1.3. Measures  

The independent variables, emotion in the picture and source of the content, were scored beforehand and the dependent variables were collected and measured through the experimental survey (see Appendix A).

(34)

The independent variable, sentiment score in the picture, is operationalized through the averages of the APN sentiment scores, which means that a mean value of emotions exists for every picture. These were rated at a 4-point scale (from -2 to 2) and were rescaled to fit the 7-point Likert scale of the emotional response. This step was important to make a comparison of both variables possible. As the survey software randomly assigns respondents to different experiment groups, the mean sentiment score of every group was calculated. The source of content was operationalized on two levels, 0 for UGC and 1 for FGC. The same operationalization was executed for the valence of emotions, the sentiment score before rescaling clearly indicated whether the sentiment score was above (positive) or below (negative) zero, and were thus classified using 0 for negative and 1 for positive.

The dependent variables in this experiment are the emotional response and the attitude towards the brand.

The operationalization and measurement for the emotional response is oriented on the study of Ad-evoked emotions by the studies of Edell & Burke (1987; 1989). The only difference is that the emotions used are based on the emotion model of Plutchik (2001) because SentiBank builds on a psychological concept and its classification of emotions. It is important for the analysis and the following measurement that the emotions, identified by the visual sentiment approach, on the Instagram postings match the emotional rating in the experimental survey. Asking respondents how the visual content made them feel can identify the emotional response on a 7-point Likert scale using Plutchik’s (2001) emotions. After being exposed to visual content, the respondent has to rate how the content made him/her feel by filling out a Likert scale for each of the eight primary emotions of Plutchik’s emotional model (anticipation, joy, trust, fear, surprise, sadness, disgust, anger). Repeated exposure to visual content may change the emotional effect and reaction the brand. Edell and Burke (1989)

(35)

allocated groups to one, five or ten time exposures and find a steady effect of feelings over time. In addition, an affective reaction to ads should occur more distinctly after three to five exposures (Petty & Cacioppo, 1983) and exposure over three times may be sufficient for consumers to acquire information about brands (Krugman, 1972). A highly increased exposure level can cause a lower efficiency of ads over time (Burke & Edell, 1986; Calder & Sternthal, 1980). Therefore this study uses five exposures to measure emotional response focusing on the goal to provoke an affective reaction and to permit respondents to generate information about the particular brand without creating risk of excessive exposure. This construct enables the measurement of positive as well as negative emotional response separately, by taking the means of the four positive and the four negative emotions after each exposure. In addition, a total emotional response score can be calculated by subtracting negative emotional response from positive emotional response. Hereby the valence of the emotional response can be detected and each value above zero is positive, each value below zero is negative and a value equal to zero is a neutral response. The same scale as for emotions on the picture is used, namely 0 for negative, 1 for positive and 2 for neutral.

The attitude towards the brand will be measured according to the scale used by Holbrook and Barta (1987), which reported internal consistency of 0.98. Respondents will be asked to evaluate the brand by indicating their tendency towards five different adjectives on a five-point scale (good/bad, like/dislike, favourable/unfavourable, positive/negative). The evaluation of the brand will be measured by taking the mean of every element.

Brand familiarity indicates the respondents’ level of awareness and knowledge about the brand. Its measurement can be included as a control variable; it may have an influence on brand evaluation. The latter may differ for brands that are not equally well known in different

(36)

countries; e.g. SuitSupply is a Dutch brand and therefore more famous in the Netherlands. While controlling for brand familiarity, a possible country-of-origin effect can be detected. The scale is developed corresponding to Zhou, Yang and Hui (2010) where three semantic differentials are evaluated on a five-point Likert scale (unfamiliar/familiar, very knowledgeable/not at all knowledgeable, have seen many advertisements/have never seen advertisements about this brand) (Zhou, Yang, & Hui, 2010).

The level of consumer involvement with the brand based on their expressed interest in the brand is measured using the scale of Rodgers and Schneider (1993), which has three Likert-scale statements. The statements (I attach great importance to this brand, one can say that this brand interests me a lot, this brand is a topic which leaves me totally indifferent) can be rated on a five-point Likert scale ranging from “strongly agree” to “strongly disagree” (Rodgers & Schneider, 1993).

A manipulation check is included in the questionnaire to ensure that the experiment is working, meaning that the respondents realize that the visuals are brand-related and notice that the content posted is either user- or firm-generated. Respondents need to indicate not only whether they saw a brand in the pictures but also which one. Followed by a question about the source of the postings.

In addition, with the purpose too screen out click-troughs, an attention-screening question was introduced in the middle of the survey and sent to respondents towards the end of the survey if not answered with the predetermined answer. This section included a text above the Likert scale, explaining the purpose of this question and made it clear for respondents, who read the description, not to be confused and only answer with the predetermined response.

(37)

At last, some demographics, like age, gender, education and country of origin, are requested as well as information on Instagram usage, which enables the possibility to compile a respondent’s profile.

4.2. Data collection  

Data is collected through an online experimental survey that is created with the online software Qualtrics, which enables to compile a survey with the required features for an experimental survey. Questions were established and data was inserted as well as the randomization of the different treatments was ensured. Respondents were distributed into sixteen different groups, changing in valence of the emotions, the source of the content and the depicted brand (see Appendix B).

4.2.1. Pilot study

As a first step, a pilot study was conducted to evaluate the study design and to check if the survey works properly and all the variables that needed for the analysis are measured. The testing phase is crucial in order to test the survey before the launch to a greater public, ensuring that there are no mistakes or misunderstandings in the questions. The respondents were asked to state whether questions were unclear or not functioning properly.

The pilot study was conducted with 19 respondents (37% male, 63% female) mainly living in Austria, Germany, Luxemburg and the Netherlands, majority with a completed Bachelor’s degree (73.7%) and aged between 22 and 28 years. The manipulation check was reviewed to check whether the experiment is able to measure the differences between firm and user-generated post. A majority of respondents (63.2%) did see a brand name in the pictures and there is a significant (p<0.01) correlation between the actual source and the source indicated by the respondents. Only small changes needed to be made to ensure that e.g. some questions are only displayed if a condition is fulfilled and spelling mistakes were corrected. The

(38)

sentiment “anticipated” was unclear for some respondents, so it was decided to include a small definition to avoid language bias in the main study.

Furthermore, the pictures and the valence rated by the Sentibank were not accurate for every picture, especially pictures classified into the negative emotions treatment group could be considered as positive on a subjective basis. This could be related to the fact that the APN method is still very unexplored and that the method may have difficulties to recognize controversy. As the selected pictures are the focus of this study, they need to be selected carefully and should permit to measure the purpose of this research.

The pictures are therefore reselected and recoded on a more subjective basis but allowing to include controversy and negative emotions to play a more significant role. The pictures were ranked again according to the number of likes and positive and negative images were chosen starting from top to bottom. A total of 96 pictures were downloaded for further analysis. Two Marketing experts with a background in Instagram usage and Online Marketing then recoded these pictures on a 7-point Likert scale on the basis of the 8 emotions of Plutchnik. The experts coded the emotions according to what emotions are depicted on the picture by taking into account facial expressions, controversy and the overall mood of the picture. 80 pictures were chosen for the main study, in which 35 were different from the first batch. To prove the coherence of the coders’ ratings, an inter-rater reliability (IRR) test is essential. As the construct is at an interval level and rated by two coders, an intra-class correlation (ICC) is the most suitable choice and is executed for every treatment group separately. IRR was determined by using a two-way mixed, consistency, average-measures ICC (Hallgren, 2012). The results varied between 0.334 and 0.878 for the four treatment groups (posUGC ICC=0.669; negUGC ICC=0.612; negFGC ICC=0.334; posFGC ICC=0.878). The values are classified as poor for negative firm-generated content, good for the positive and negative

Referenties

GERELATEERDE DOCUMENTEN

Chapter 5 explored the sub questions, providing a basis to answer the overarching research question, “How do the forms of upgrading present in Cambodian social enterprises (SEs)

The purpose of this study was to make a comparison between two European markets – the Western, represented by the Dutch consumers and the Eastern, represented by the

The study discovered that, in addition to not responding to some employees’ PDPs, the Research Unit rejected the application of some employees, who wanted to attend training outside

Considering the advantage of the baseline over the simplest Votes method and that the baseline is one of the most ef- fective methods known, we may conclude that the improve- ments

The results show that the proposed seed extraction mechanism derive random seed from sensor data and the seed is capable of passing 12 out of 15 NIST statistical tests.. The results

The goal of this study was to discover Implicit Leadership Theory in Indonesia by exploring leadership prototype based on perception of Indonesian

The second table will present the frameworks, models, and theories identified by name; categorization (framework, theory, or model); studies and projects that applied it; approach

Over 24 h, chiral amplification is apparent in water/acetonitrile and is translated into a gradual increase of CD signal, which indicates that the chiral and achiral building blocks