Increasing box-office and online revenues : the effective utilization of buzz and word-of-mouth marketing

(1)

Master thesis

MSc in Business studies

Entrepreneurship and management in the creative industries Increasing box-office and online revenues: The effective utilization of buzz and word-of-mouth marketing

Supervisor: Frederik Situmeang Second supervisor: Joris Ebbers Name: Xiomara Blank

(2)

Statement of originality

This document is written by Student Xiomara Thalita Blank, 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)

Acknowledgements

This thesis was written as the last major assignment to achieve the Business Administration specialization Entrepreneurship and Management in the creative industries Master degree. The process was longer than expected; however the help of Frederik Situmeang, Maarten van Duren, Rahul Mohan, Vanessa Belfor, Rudolf Blank, and Margo Blank van Els made the journey feel shorter.

First and foremost I want to thank Frederik Situmeang. Your guidance during this process has kept me from having several meltdowns, when R was not doing what I desired. Second, thank you, for your hints and guidance with R, Maarten van Duren. During the last stages of creating the codes needed in R Rahul Mohan patiently helped my resolve data storing issues. Thank you so much for your swift and kind help. Last but certainly not least, I want to thank my parents Margo Blank van Els and Rudolf Blank, and my sister Vanessa Belfor. Even though the process was long and hard, your support and kind words have helped me reach the finish line.

(4)

Abstract

The purpose of the present study was to research the effect of eWOM (e.g. sharing positive or negative product or service information/evaluations from consumer to consumer (Daugherty & Hoffman, 2013). through social media platforms in regards to the motion picture industry. WOM has been found to significantly affect firm outcomes in regards to financial and market positioning success (Libai, Bolton, Bugel, Ruyter, Gots, Risselada & Stephen, 2010, p. 277). Many aspects of WOM have been researched, yet the effect of WOM in the realm of channel characteristics has been identified as a gap in the literature by Libai et al. (2010).

The present study focuses on the financial success of movies promoted through social media micro-blogging platforms, such as Twitter. Financial and descriptive data on movies released in 2010 - 2012 was gathered from various movie databases. Tests were run to establish whether sentiment scores have a mediating effect on the relationship between production costs and revenues. Results show a statistically significant mediating effect of sentiment scores on box-office and after sales revenues. Thus, targeted marketing through social media micro-blogging platforms aimed at improving the sentiment about movies can help improve the spread of positive eWOM among consumers, ultimately leading to increased box-office and after sales revenues.

The present study has the following limitations. First, sentiment on movies was established through tweets with hashtags related to the movie title. However, hashtags are not only used to tweet about the movie, but also for tweets about topics other than the movie. Second, acrimony was not considered during analysis, which means the analysis did not take sarcasm into consideration during the scoring of tweets. Third, emoticons that could not be translated were removed, which means that

(6)

a layer of sentiment was considered. Fourth, re-tweets were not removed from the sample, which may have resulted in extra weight given to such tweets.

(7)

1. Introduction

A growing group of consumers does not visit movie theatres; instead, they stay home to enjoy their “new technologies and advanced personal theatre systems”. This is “the revolution in digital home entertainment” (Mohr, 2007, p. 396). Another important trend is a revolution in the distribution channels, which involves Internet film

distribution (e.g. streaming) and digital cinema (Silver & Alpert, 2003). Trends such as these create a marketing challenge for movie distributors. Currently, studios market motion pictures through the use of celebrities in their advertisement campaigns to utilize star power and create buzz. However, it has been shown that star power does not predict box office revenues. Volume of WOM (e.g. sharing positive or negative product or service information/evaluations from consumer to consumer (Daugherty & Hoffman, 2013) either online or offline, which is generated whenever movie studios target audiences and infiltrate the first stage of consumer decision-making (i.e.

awareness, through services such as micro-blogging), can predict box-office revenues (Liu, 2006). There is an increased interest in customer-to-customer (C2C) interactions (e.g. WOM) because these can potentially “influence a firm’s financial performance and market position” (Libai, Bolton, Bugel, Ruyter, Gots, Risselada & Stephen, 2010, p. 277).

Generally, WOM studies focus on the antecedence and/or effects (Liu, 2006, Thurau, Gwinner, Walsh & Gremler, 2004, Trusov, Bucklin & Pauwels, 2009). Other studies shed light on long-term performance (Dhar, Sun & Weinberg, 2012), and box-office revenues (Liu, 2006, Basuroy & Chatterjee, 2008). Kimmel and Kitchen (2013) call for a better understanding of eWOM communication and its impact. In their paper (2010) on C2C interaction, Libai et al. ask if channel characteristics “moderate the effect of C2C interactions on purchase behaviour” (p.274). This thesis focuses on the

(8)

use of social media networks as a channel to stimulate WOM, in order to increase theatre ticket sales and after-theatre sales – thus addressing the gap identified by Libai et al. (2010). In other words, this study looks at the role that WOM plays on social media platforms in the successful exploitation of movies and will focus on the following question: “How can movie distribution studios effectively utilize buzz and word-of-mouth in order to increase box-office and after-theatre revenues of released movies?” Answering this question will add to our understanding of eWOM and its effective utilization as a marketing tool. The study is based on a literature review and an empirical experiment. The empirical setting of this study is the American movie industry. This industry is characterized by its oligopolistic nature (i.e. there is a small number of firms that supply products that are close substitutes, and any pricing and supply decisions of competitors influence the focal firm). Further characteristics include large numbers of content creators and a small number of distributors (Vogel, 2001). During this study, the focus is on the Hollywood majors (i.e. Disney, Fox, MGM, Paramount, Sony pictures, Columbia, Universal and Warners) and smaller movie distribution studies. The methods used during empirical testing relate to movies released between 2010 and 2012 and consists of quantitative numerical ratio data retrieved from movie databases ‘the-numbers’ and ‘box office mojo’. Data in regards to social media platforms was collected using Twitter data. Analysis of the presumed mediating effect was tested using the Baron & Kenny (1986) method.

The thesis is structured as follows. In the first section, the conceptual framework is described. The second section presents a review of the contemporary literature. In the third, the research design and methodology are specified, and the Baron & Kenny (1986) method is further elaborated upon. In the fourth section of the thesis, the empirical data is analysed. In the fifth section, the results of the tested hypotheses are

(9)

illustrated. In the sixth section of this paper, the discussion is construed and managerial implications are reported. The final section concludes the thesis.

(10)

2. Conceptual framework

Selection system theory

Selection system theory argues that the market consists of “selected” and “selectors”. The selected are actors competing with each other, whereas selectors are the actors who determine how valuable the offered goods are. There are three types of selection systems: market, peer and expert. Consumers who assign value to the product

construct a market selection system. The peer selection system entails producers providing consumers with information, which the consumers in turn will use to estimate how valuable the product is. In an expert selection system, actors who are neither producers nor consumers, but who have particular knowledge and expertise, assign value to products. (Mol & Wijnberg, 2007)

In the context of the movie industry, the studios are producers, who first evaluate the value of the movie and then (if it is estimated as valuable and projected to generate an acceptable return) distribute the movie. This points to a peer selection system; however, after the movies have been released to the public, consumers are the ones who assign value to the movie. Consumers assign value to the movie based on the expert opinions of movie critics and their environment (e.g. friends/family), which can be argued to be a form of value creation through WOM. The assigned value determines the overall revenue stream and potential value of a franchise based on the original release.

Previously, movie studios assumed that the stars are leading criteria during value creation. However, as mentioned above, Liu (2006) shows that star power is an ineffective box-office revenue predictor, and the volume of WOM is an effective predictor. Based on selection system theory, there should be more focus on the value

(11)

judgement of consumers, since consumers enable eWOM and other similar consumers trust their value judgements.

Similarity attraction/social categorization

The social categorization perspective describes how actors distinguish between in-group members and those who do not belong in the in-group. The focus is on

interpersonal similarities, which determine interpersonal attraction. For example, actors make distinctions based on demographics, ideals, perspectives, values, orientation, perspectives and attitude (Van Knippenberg & Schippers, 2007). Based on this theory, it can be argued that consumers will surround themselves with actors who are similar to them (which infers similar interest and value judgements) during the consumer decision-making process.

Consumer decision-making theory

Consumer decision-making theory argues that consumer decision-making takes place in stages. The three main consumer decision-making stages are: awareness,

consideration and actual choice. During the selection process, the set of products or services becomes smaller as the consumer passes through the stages. The first criterion of the decision-making process is consumers’ awareness of a product or service, before they can consider alternatives and finally make an actual choice.

In the first stage (awareness), consumers become aware of the producers and their products or services. This awareness creation takes place through advertising, WOM or any other form of exposure to the product or service resulting from the marketing activities of producers. During the second stage (consideration), consumers

(12)

look for alternative products or services that fit their interests. During the first two stages, products and services are considered and rejected, and it is in the third and final stage (actual choice) that consumers actually choose a product to purchase or service to consume (Kuijken, Leenders, Wijnberg & Gemser, 2013).

Movie studios can create consumer awareness through social media presence and advertising. Since actors have the tendency to befriend similar people, it can be argued that their friends on social media will have similar interests. This means that actors who are aware of the product can spread the news of this product to those similar to them, creating a higher-level awareness in their environment. A

consequence of this sharing of information among actors could result in a decline in marketing expenditures and more targeted marketing for movie studios, since

consumers themselves would spread the news about the movie among similar others, instead of distribution studios creating expensive marketing campaigns.

(13)

3. Literature review

Buzz and WOM

Word-of-mouth (WOM) was first formally studied in 1955 (Brown & Reingen, 1987). “Classic WOM research has typically studied dyads, consisting of a receiver and a sender” (Libai et al. 2010). In the literature, WOM is also referred to as buzz; these terms have been defined in a variety of ways. WOM, according to Liu (2006),

“involves informal communication among consumers about product and services” (p. 74). According to Mohr (2003), WOM is “the process by which an individual

influences the actions or attitudes of others” (p. 396), while Kariouchina (2011) defines the term as “consumer excitement, interest and communication around a project”. Consistent with the WOM definition of Mohr (2003), in the context of the movie industry, actors who practise WOM can influence the purchasing behaviour and attitudes of the receivers towards movies. WOM can emerge on a macro- and micro-level. According to Brown & Reingen (1987), WOM on a macro-level involves cross group communication, whereas micro-level WOM consists of dyads or small groups. They go on to argue that setting boundaries on the concept of WOM only creates the appearance of solving the reality of the open systems that characterizes consumer behaviour.

In the present study, WOM is constrained by personal relations among consumers (1987) and their behaviour on the Internet. In this study, the words ‘buzz’ and ‘WOM’ are used interchangeably, and are defined as sharing positive or negative product or service information/evaluations from consumer to consumer (Daugherty & Hoffman, 2013) either online or offline. The present study focuses on Internet-based

(14)

or online WOM, frequently referred to as electronic WOM or eWOM (Thurau et al. 2004). The online environment enables a new form of conversation in and among groups, consisting of a large numbers of participants who take on different roles (Libai et al. 2010). Participants of online communities, in the context of movies, can perform the role of the advisor or reviewer/critic, which indicates a “market-expert selection system”, with the term ‘market’ referring to the consumers performing the role of a movie critic and the ‘expert’ referring to legitimacy of set expertise, being gained through other consumers recognizing the expertise on the subject.

Recognition, however, can be achieved with or without official credentials, thus the role is suitable for actors whose WOM is trusted to be integrated with knowledge of what makes good or bad movies. Previous studies (e.g Eliashberg & Shugan, 1997) found a predictor effect for critics on the aggregate box-office level. In addition, Holbrook & Addis (2007) show that consumers have good taste, in other words they found a significant relationship between expert judgment and popular appeal,

supporting the market-expert selection system as on of the underlying assumptions of the hypothesis.

WOM is either organic or amplified. Organic WOM “occurs naturally, without the firm’s intervention”, whereas amplified WOM consists of “campaigns designed to encourage or accelerate WOM” (p. 270). WOM behaviour is a continuous variable, which consists of two components: relation content and relational form (Brown & Reingen, 1987). “Relational content refers to the substantive type of relation represented in the connection among individuals” and “relational form refers to properties of the linkage between pairs of actors that exists independently of specific contents” (1987, p. 351).

(15)

The nature of the “recommender-receiver relationship” (Ryu & Feick, 2007, p. 85) (e.g. relational form) can be either strong or weak. The strength of a tie is

determined based on a linear combination of time, emotional density, intimacy, and reciprocal services, with each of these characteristics being independent of each other but also being highly inter-correlated (Granovetter, 1973). A strong tie is a person with familial connections or who is considered a (trusted) friend, whereas weak ties are people who are described as loose friends or acquaintances (De Meo, Fiumara & Provetti, 2012). The hypothesis that allows for the extrapolation of dyadic ties to larger structures is as follows; as the strength of the tie between A and B grows, the larger the proportion of individuals in the social network to whom they will both be connected by a strong or weak tie. Empirical evidence further proves the plausibility of the hypothesis, since there is evidence that the stronger the tie, the more similar individuals are on other aspects (Granovetter, 1973). Thus, strong ties are established with actors who are similar, supporting the similarity attraction/social categorization theory. Strong ties are argued to play an influential role (whereas weak ties play a conducive role) during information flows. Weak ties generate macro WOM and strong ties micro WOM (Brown & Reingen, 1987).

WOM can be positive or negative. Consumers are motivated to engage in positive WOM by “altruism, product involvement”, self-enhancement (Thurau et al., 2004, p. 40), and when they want to help the company. Negative WOM behaviour displayed by consumers has the following motives: “altruism, anxiety reduction, vengeance, and advice seeking” (Thurau et al., 2004, p. 40). However, in the film industry, Liu (2006) has found that valence (i.e. recommendations and assessments) is of less importance when it comes to predicting box office revenues than the amount of WOM. According to Liu (2006), the explanatory power of WOM comes from its

(16)

volume, which causes him to call for more targeted advertising. In order to achieve targeted marketing, movie studios can utilize collaborative filtering systems, as identified by Gladwell (2000) to potentially cause the irrelevance of blockbusters (i.e. products that enjoy fame and heavy marketing) and to create sleepers (i.e. products that slowly come to the attention of customers), in the form of online social media networks. These networks can be utilized since the social media platforms accumulate information about their users, which in turn can be used during targeted marketing. Consumers are inclined to ask similar other people for product information, and when they are looking for relational goods, they are more likely to activate a strong tie to gather information (Brown & Reingen, 1987). It is further argued by Brown &

Reingen (1987) that consumers that are part of the same group are more likely to have similar preferences, another argument for the use of social media networks during marketing efforts of movie studios.

The movie industry has a short life cycle, combined with frequent introduction of new movie releases (Krider & Weinberg, 1998). Each year, more movies are being released. However, despite the fact that sequels perform more favourably, the number of released sequels has remained constant (Dhar, Sun & Weinberg, 2011). According to Elberse & Eliashberg (2003), two dimensions of competition for screen can be distinguished, “new releases” versus “ongoing movies”. Consumers favour sequels, as is reflected in their performances, and this demand in turn creates the possibility of their establishment.

Before the release of a movie, audiences have high expectations and it is this period in which WOM is at its peak (Liu, 2006). Consumers tend to rely on customer-to-customer interaction (i.e. WOM) when they are less aware or lack knowledge about a product (Libai et al. 2010). According to Karniouchina (2011), star buzz (i.e.

(17)

buzz about a known actor/celebrity) also has a positive influence on anticipation and box office revenues. However, after the release, audiences tend to lose interest (Liu, 2006), which leads to a decrease in WOM volume, and which in turn explains the shrinkage in box office revenues. In addition, studios engage in active marketing activities before the release of a movie, since exhibitor fees go up as the length of time in the theatre continues. Which means less signals are being sent to consumers in the weeks following the theatre release of the movie. Yet in subsequent weeks, WOM is found to be an important predictor of box office revenues and allocated screens (Elberse & Eliashberg, 2003). This confirms the finding of Liu (2006) (i.e. the volume of WOM is an important box office revenue predictor).

According to Thurau et al (2004), our understanding of what motivates consumers to engage in WOM activities can be used to utilize marketing strategies that drive consumers to WOM behaviour. They found that consumers are primarily motivated to engage in WOM by the desire for social interaction, economic

incentives, “their concern for other consumers, and the potential to enhance their own self-worth” (Thurau et al, 2004, p. 39). Consumers who want to use movies as

identity goods (i.e. consumption of goods to generate an expression of how the consumer wants to be perceived by others) are motivated by their desire to enhance their self-worth. Through the use of social media networks, it can be argued that consumers will be motivated by their desire of social interaction to employ WOM behaviour. Movie studios can save on global marketing costs through WOM, since contemporary marketing costs can exceed production costs (Silver & Alpert, 2003). Besides lowering costs, WOM strategies increase delivery speed and overcome consumer resistance (Trusov, Bucklin & Pauwels, 2009).

(18)

Through the release of movies, distribution studios send quality signals about their existing and future products to consumers. Credible quality signals will lead to an increase in the quality perception of the movies. Effective utilization of WOM will lead to an increase in the perception of quality, due to consumers’ confidence that information gathered through WOM is non-biased, or paid for by the focal company. Another reason for an increase in quality perception is sequels, which have been advertised with the same or a smaller marketing budget than the original release (Busuroy, Desai & Talukdar, 2006). According to Busaroy et al. (2006), consistent with signalling theory, rational consumers trust quality signals because they assume that the focal company does not supply false information, as they know that this would be “economically unwise” (p. 288). However, they argue that their results could possibly be explained through the use of other behavioural theories. Thus, sending quality signals through social media channels will motivate consumers to create movie buzz or WOM volume. Currently, the two biggest social media networks are Facebook and Twitter (eBizMBA rank, 2014). It is expected that movies that have positive eWOM on social media will be associated with high box office revenues.

Collaborative Filtering Systems

At the root of collaborative filtering systems is basic human behaviour: sharing of opinions (i.e. WOM). Through filtering or evaluating the opinions of people about a certain item (i.e. product or service), the system takes on its shape. A collaborative filtering system can give recommendations or make predictions of a consumer’s opinion. The item can be anything a consumer can give a rating for. These ratings come in many different forms (scalar/ordinal, binary and unary), and can be gathered explicitly and implicitly. Explicit gathering of ratings consists of a consumer giving a

(19)

rating of an item. Ratings that are gathered in an implicit way are based on the consumers past actions. Collaborative filtering systems can provide consumers with advice about a certain item, help them find (new) items they might like, and help them with tasks that are domain-oriented (Schafer, Frankowski, Herlocker, & Sen, 2007).

The collaborative filtering systems are currently employed by websites such as Amazon.com and Netflix to recommend products or movies to their target audiences. Currently, social media websites tend to recommend advertisements to users who have similar other ‘friends’ who liked or rated a page, which can be categorized as implicit data-gathering of the social media platforms. Collaborative filtering systems on social media thus can be argued to be a form of market selection, since it is the consumer who does the quality rating. These ratings in turn determine the value that is prescribed to the movie, when it is recommended to other (similar) consumers.

Collaborative filtering systems are useful for movie distribution studios because of their recommendation function. Consumers who are unaware of a newly released movie that is currently in theatres cannot enter the consideration stage, which means they cannot enter the stage of actual choice during the consumer decision-making process. However, when a consumer receives a recommendation, based on ratings of their friends (i.e. similar others), this will launch them into the awareness stage. The perceived/prescribed quality of the movie will then determine the stage the consumer will reach in the decision making process. Thus, once a consumer has been made aware of the existence of a movie, the perceived quality of the movie and statements made by their friends about the movie will be the decisive factors for the consumer to decide if they will purchase a theatre ticket for the movie or purchase DVDs after theatre showings when they enter the stage of actual choice.

(20)

Social media platforms offer companies the opportunity to use their advertising services, which allow companies to accurately market to their target audiences. This in turn creates a cost-effective way to reach the desired audience. The techniques used to locate a specific type of social media user most likely to be

interested in the advertisement can be argued to originate from techniques employed by the collaborative filtering systems.

Social Media Platforms

On social media platforms, consumers read and share their opinions, evaluations and factual information on a broad range of topics. The present study focuses on the subjective expressions of consumers in regards to movies first released in theatres. Content on online communities (thus also on social media networks) is mostly user-generated (Trusov, Bucklin & Pauwels, 2009), while another section of the content is generated by targeted advertising (e.g. collaborative filtering systems are employed). According to Trusov, Bucklin & Pauwels (2009), a natural form of spreading

information in online communities is through the use of eWOM.

Contemporary social media platforms allow users to share content (which they have found while surfing the web or on the same or another social media platform) with their friends. Moreover, consumers have the opportunity to give their opinions on the posts they shared or posts that their friends shared with them.

In the context of social media, connections among social media users can be categorized as weak ties, since these connections are between “individuals who otherwise belong to distant areas of the friendship graph” (De Meo, et al., 2012, p. 2). According to Granovetter (1973), there is always a strong or weak tie between node “A, B, and any arbitrarily chosen friend of either or both”(p. 1363). He goes on to

(21)

argue that a “bridge” is the only line in the network that signifies a path between two nodes, concluding that “all bridges are weak ties” (p. 1364) (i.e. a larger number of actors can be reached through a weak tie when anything needs to be diffused), due to their ability to more effectively bridge social distance. He argues that transitivity is a function of the strength of the ties, which leads weak ties to be more likely to connect actors of different groups (Granovetter, 1973). However, De Meo et al (2012), caution that not all bridges are suitable for analysis of large-scale social networks, such as Facebook and Twitter, due to the “small world effect” (i.e. short paths that connects any two nodes) and “scale free degree distribution” (i.e. hubs that maintain efficient connection within the network), since the likeliness of finding a node (of which deletion will lead to an inability of two nodes to connect through alternative paths) is very low. To solve this problem, they define “shortcut bridges” as the path between two nodes, that if deleted would create increased distance between them. A limitation of the definition of shortcut bridges is the high computation costs of the pairs with the shortest paths, and the arbitrariness of the concept of distance between the nodes (De Meo et al, 2012).

Weak ties, in the context of social media networks, are thus defined as ties that connect pairs of nodes belonging to different communities. Two important concepts regarding weak ties are (i) the symmetry of the relationship of the two nodes and (ii) the weight assigned to the connection. The symmetry of the relationship speaks to the non-directionality of the relationship (i.e. mutual friendship), which is the case for Facebook friends. It is argued by Libai et al (2010) that the structure and dynamic of a social network has a fundamental effect on WOM and firm outcomes. According to De Meo et al. (2012), Twitter has hierarchical connections, while Facebook reflects a friendship social structure. According to Pak & Paroubek (2010), due to the free

(22)

format and easy accessibility of micro-blogging platforms, consumers on the Internet tend to shift traditional communication tools to micro-blogging services, such as Twitter, Tumblr, and Facebook (p. 1320). In their 2009 paper, Jansen, Zhang, Sobel, & Chowdury identify micro-blogging as a new from of eWOM, since these mediums allow people to share “brand impacting thoughts (i.e. sentiment)” almost everywhere and with anyone connected to the internet (p. 2). Thus, they argue that “micro-branding comments are immediate, ubiquitous, and scalable”. The social media platform Twitter will be subjected to analysis. Reasons to use Twitter during analysis include its diverse pool of people’s opinions, “the collected corpus can be arbitrarily large”, posts from different social and interest groups can be collected, and users are globally represented (Pak & Paroubek, 2010, p. 1320).

Twitter is a contemporary popular social media and micro-blogging platform with 225 million active monthly users. Each post is referred to as a Tweet. Four unique attributes of tweets have been identified by Go, Bhayani, & Huang (2009): (1) length (e.g. a maximum of 140 characters), (2) data availability (e.g. the Twitter API or Application Programming Interface), (3) language model (e.g. frequent

misspellings and use of slang), and (4) domain (e.g. a wide variety of topics are discussed) (p. 2). The rationale behind a limited amount of characters is to keep the message grounded in the present and focused on the topic at hand (Twitter, 2014). According to Pak & Paroubek (2010), tweets vary from “personal thoughts to public statements” (p. 1320). Twitter allows users to create a personal profile, which results in their personal Twitter name (visualised by a “@” followed by their username). Each tweet can contain links to videos, articles or hashtags. The Twitter hashtag, symbolized by “#”, represents a link to the topic discussed in the short message, which allows users to search a hashtag and get insight on all tweets containing the

(23)

hashtag of interest (Twitter, 2014). According to Cunha, Magno, Comarela, Almeida, Gonçalves & Benevenuto (2011), “hashtags are used in Twitter to classify messages, propagate ideas and also to promote specific topics and people” (p. 58). It is argued by Agarwal, Xie, Vovsha, Rambow, & Passonneau (2011) that consumers use hashtags to mark topics and increase the visibility of their message (i.e. tweet).

Twitter allows companies to advertise their service(s)/brand(s)/product(s) they offer, through promoted accounts, tweets, and trends. The promoted account allows companies to target specific users who are more likely to follow the company. Likewise, promoted tweets allow companies to target their tweets to a specific audience, based on (for example) demographics, location, interests, and keywords. Twitter in turn charges companies based on the amount of “retweets”, favourite tweets, or clicks on the promotion tweets (Twitter, 2014).

The growth of micro-blogging social media platforms, such as Twitter, has lead organisations to seek ways to mine Twitter information to establish

understanding of the consumers’ opinion (Kouloumpis, Wilson, & Moore, 2011). These authors also argue that sentence level sentiment analysis is most similar to establishing sentiment of tweets (p. 538). Nasukawa & Yi (2003) argue that the identification of sentiment in expressions is more reliable than overall opinions (p. 70). Social mediating technologies “have the potential to substantially impact word-of-mouth branding” (Jansen et al. 2009, p.1).

(24)

Sentiment analysis of micro-blogs

Sentiment has been defined by Go et al. (2009) as a subjective positive or negative feeling (p. 2). A main subject in sentiment analysis is the identification of the way sentiment is expressed in texts and if these expressions are positive (favourable) or negative (unfavourable) opinions (Nasukawa & Yi, 2003, p.70). They go on to identify three parts of sentiment analysis: (1) expression of sentiment, (2) the polarity and strength of expressions, and (3) the relationship of the expressions to the subject (p. 71). During the current study, sentiment is defined as positive or negative opinions towards movies.

Earlier research has implemented sentiment analysis through tweets. For example, Go et al. (2009) employ distant supervision techniques to train data

consisting of tweets with emoticons, which are either positive or negative (e.g. “:)” or “:(“ (p.1). In their 2003 research, Nasukawa & Yi identify sentiment based on

fragments of texts about a subject within documents (p. 71). In addition, they argue that sentiment analysis techniques can have a great impact on “competitive analysis, marketing analysis, and detection of unfavourable rumours for risk management” (p.70). In their 2009 research, Jansen et al (2009) found that micro-blogging can be a powerful tool to gain competitive intelligence and a way to differentiate the focal organisation from its competitors (p. 6). The 2009 paper of Jansen et al considers the impact of social mediating technologies (SMTs) on companies; specifically they look at the customer relationship. They identify branding as a key element in customer relationship (p. 1). They go on to argue that elements that generate the power for eWOM branding are its immediacy, significant reach, credibility, and accessibility (p. 2).

(25)

WOM theory argues that consumers will engage in positive WOM under certain conditions (i.e. altruism, product involvement, and self-enhancement). According to Jansen et al. (2009), the main motivator to spread positive eWOM is to gain “social or self-approval”, since positive eWOM demonstrates the senders’ “splendid purchase decisions” (p. 6). These motivating conditions combined with micro-blogging services, allow consumers to position themselves as movie experts. Strong ties to these consumer movie experts are established through temporal, constant, and positive interaction, leading to increased eWOM and biasing the consumer decision-making process, since products and services offered by a perceived similar recommender are more likely to enter the awareness stage during the consumer decision-making process.

Jansen et al. (2009) identify the need for specialised marketing efforts and methodologies to analyse the sentiment of posts on micro-blogs, such as Twitter (p. 9). The majority of micro-blogs contain seeking or sharing of branding comments (p. 1). They also emphasise the importance of understanding how micro-blogging is changing eWOM branding, since these changes can have a significant impact on “the success of advertises, businesses, and products” (p. 2). An important part of brand management is understanding the opinions of consumers about promoted brands and products (p. 6) Due to daily growth of the micro-blogging platforms and services, data available for sentiment analysis and opinion mining also grows (Pak & Paroubek, 2010, p. 1320).

(26)

The following is hypothesised during the present study: sentiment scores of successful movies are higher than that of less-successful movies. Following this logic, it is expected that the sentiment score of a movie mediates the relationship between the production budget and box office, leading to the following hypotheses:

H0: Sentiment score does not mediate the relationship between production budget and box office revenues

H1: Sentiment score mediates the relationship between the production budget and box office revenues

The same assumption is made of the influence of sentiment scores during after sales, thus, sentiment scores of movies with successful after sales will are higher than that of less-successful movies. Leading to the following hypotheses:

H0: Sentiment score does not mediate the relationship between the production budget and after sales

H1: Sentiment score mediates the relationship between the production budget and after sales

(27)

4. Research design and methodology

Research setting

The context of this study is the American movie industry. The sample consists of 296 movies ranging from the year 2010 to 2012. This sample consists of new releases and franchises released and distributed by the previously mentioned major and smaller movie distribution studios.

Data collection

Quantitative numerical ratio data, such as budgets and revenues, on the movies in the sample is gathered from ‘the numbers’ database. The-numbers database is used due to its size and usage in previous studies (i.e. Qin, 2011). Consistent with the Qin (2011), Dhar et al (2011), and Duan, Gu & Whinston (2008) study, another movie database used to gather data from was ‘Box-office mojo’.

Movie presence on Twitter in the form of a hashtag was established through the twitter search engine, with the search term “movie title”. Movies with frequently used words were searched for twice, first with the aforementioned search term, and if necessary also with the search term “movie title + movie”. In the case of movies which are franchise and were not found through the general search term, these movies were then searched using the term “franchise + title movie”. Tweets with a hashtag, matching movie titles, were collected through R.

R is considered “a powerful language and environment for statistical computing and graphics” (Torfs & Brauer, 2014, p.1). The language is an

(28)

2002, p. 2), mostly used as a research and educational language (Torfs & Brauer, 2014). R was used to create numeric lingual sentiment indicators, which were then used to further analyse the relationship between eWOM and (box-office) revenues with SPSS 21.In their 2011 paper, Agarwal et al find that standard natural language processing tools are useful, regardless of the genre in which the tool was trained (p.30). The importance of natural language processing during the identification of sentiment expressions, and analysing of the semantic relationship of the expression with the subject is emphasised by Nasukawa & Yi (2003, p. 73). The R environment was initially set up using the functions described in Appendix A1. The packages used were devtools, twitteR, ggplot2, rJava, xlsx, plyr, stringr and MASS. According to Go et al. (2009), the twitter API “twitteR” provides an easy way to “extract large amounts of tweets” (p. 1). The Hu Liu lexicon (University of Illinois Computer Science, 2015), consisting of 2006 positive and 4783 negative words, was used to score the sentiment for each tweet individually. The scores were established using the algorithm created by Breen (2011), which was initially used to analyse sentiment of airline industry data, also see the function “Sentiment analysis” in appendix A1. The sentiment analysis function “implements a very simple algorithm to sentiment, assigning an integer score by subtracting the occurrences of negative words from that of positive words” (Github, 2015). According to Nasukawa & Yi (2003), quantitive sentiment analysis can be reached through counting positive and negative sentiments (p. 76), supporting the use of the Breen (2011) algorithm. The “clean emoticons” function, also see appendix A1, removes all punctuation and numbers. In their 2009 paper, Go et al call for the utilisation of emoticon data during sentiment analysis (p. 6). The current study solved this through the translation of emoticons. “Emoticon description” function, detailed in appendix A1, replaces known UNI8codes or bites with the

(29)

matching Whatsapp emoticon description listed (Timwhitlock, 2015). In their 2011 paper, Agarwal et al. define emoticons as “facial expressions pictorially represented using punctuation and letters”. However, their definition fails to include Bytes and Uni8Codes, which allow consumers to see actual graphical representations of facial expressions. Emoticon services, such as Emoji, have increased the ease and usage of emotions during messaging; which in turn has led to an increased use of emoticons in tweets. Usage of emoticons proves itself globally integrated, through consumer-requests for more diverse facial expressions (e.g. matching skin tones of diverse racial categories). According to Go et al (2009), emoticons are noisy labels due to their inability to perfectly define sentiment (p. 2). However, during the present study emoticons are translated into their matching description to improve the accuracy of sentiment scores. It should be taken into consideration that not all consumers consider the descriptions of the emoticons they use. For example, the emoticon described as loudly crying is used to represent tears of sadness, but is also used to represent tears of joy.

Data collection was then done through the use of the code formulated in Appendix A2. Agarwal et al. (2011) argue that a true sample of actual tweets is collected in a streaming fashion (p.31). Following earlier research, the last 1,500 tweets posted were collected and included in the sample. The Twitter API allows programmatically accessing tweets by query term. The Twitter API also has a

parameter for language (Go et al. 2009, p.3). All the tweets collected are set to match the English language. A container and lists for each phase in the process of data collection were created and stored. In the first phase, 1,500 tweets for each movie were collected and stored. In total, 444,000 tweets were collected. During the second phase, emoticon codes in each tweet were translated to match the emoticon and

(30)

stored. Unknown codes were removed by the emoticons function. The third phase consisted of scoring the sentiment of each tweet.

Dependent variables

Box office revenue will serve as the dependent variable in the present study, consistent with previous studies by Eliashberg & Shugan (1997) and Basuroy & Chatterjee (2008). Box-office revenue is established through two sub-variables: domestic box-office revenues and foreign box-office revenues. Another dependent variable during experiments is the after box-office Blue ray and DVD sales.

Independent variables

The production budget will serve as an independent variable during the present study. Production budget is used since the range of the budget serves as a signal from the studios about the projected revenues (or former success in the case of franchises).

Mediating and control variables

Consumer quality perception is operationalized through the sentiment score of tweets. Therefore it is assumed that consumers reflect their genuine feelings and judgements through use of specific words in the tweets related to the movie title. Consistent with Jansen et al (2011), sentiment scores are assumed to impact the relationship between production budget and (after box-office) revenues. Thus, the sentiment score will serve as a mediating variable during present study. The mediating effect is tested using the Baron & Kenny (1986) method. Franchises will serve as a control variable during the experiments.

(31)

The Baron & Kenny (1986) method

According to Baron & Kenny (1986), the mediator is the variable that accounts for (part of) the effect between the independent and dependent variable in terms of direction and/or strength. Thus, they identify 3 types of paths, of which two are causal: (1) the direct effect of the independent variable, (2) the mediating effect, and (3) the effect of the mediating and independent variables on the dependent variable. There are several conditions for a variable to function as a mediator: (a) the

independent variable has a significant effect on the variable which is presumed to be the mediator, (b) the mediator has a significant effect on the dependent variable, and (c) when the independent and mediating variables are controlled, the previously significant effect of the independent and dependent variables are non-significant. There is perfect mediation if the independent variable no longer has a significant effect when the mediator is controlled. For the use of a multiple regression during testing, two assumptions need to be met: (1) there are no measurement errors in the mediator and (2) the dependent variable is not caused by the mediating variable (Baron & Kenny, 1986).

Limitations of the Baron & Kenny (1986) method

The Baron & Kenny method does not test the indirect effect of the moderating variable. Another limitation to this method lies in the use of the multiple regression analysis, which presumes that the dependent variable does not cause the dependent variable. This means that if the mediating and dependent variable are mistakenly chosen this will lead to flawed results (Baron & Kenny, 1986).

(32)

5. Analysis

This section is a description of the analysis, with focus on the dummy variables, nominal variables, mean values, the correlation matrix, reliability and normality of the variables used. General information about the movies consisted of the following variables: Title of movie, distribution studio, franchise (dummy), genre, and MPAA rating. Word-of-Mouth (WOM) on social media platforms was operationalized through the sentiment score (N=296, M = 0,52150, SD = 0,111928) of movie hashtags. Financial information on the movie was analysed through the use of production budgets (N = 296, M = $ 54.671.804,10, SD = $ 56.956.378,90), world gross box office revenue (N = 296, M = $ 168.724.791, SD = $ 234.529.252), consisting of domestic (i.e. U.S) revenues (N = 295, M = $ 67.853.168, SD = $ 81.041.854,80) and foreign revenues (N = 285, M = $ 106.755.830, SD = $

164.242.415). After sales variables DVD sales (N = 243, M = $ 23.128.564,70, SD = $ 27.934.403,10) and Blu-ray sales (N = 224, M = $ 10.989.135,20, SD = $

15.271.671,30) are also used.

Correlation among the variables was tested using the Pearson correlations. As expected, the variables domestic (r = 0,940, n= 295, p = 0,000) and foreign revenues (r = 0,750, n= 285, p = 0,000) are significantly correlated with world gross revenues, as these are sub variables of world gross revenues. The after sales variables DVD sales (r = 0.714, n= 243, p=0.000) and Blue ray sales (r= 0,893, n = 224, p=0,000) are also significantly correlated with world gross revenues. Consistently, world gross and production budgets (r = 0,775, n = 296, p = 0,000) are significantly correlated.

Another significant correlation exists between variables blue ray sales (r = 0,737, n = 224, p = 0,000) and production budget. Similarly, domestic (r = 0,718, n = 295, p 0,000) and foreign (r = 0,750, n = 285, p = 0,000) sales have a significant correlation

(33)

with the variable production budget. The significant correlations, mentioned above, all indicate a strong significant correlation among variables as all Pearson’s r are higher than 0.5. However, a medium correlation excites between DVD sales and production budget (r = 0,490, p < 0,05). There seems to be a weak significant

correlation among sentiment score and production budget (r = 0,123, p < 0,05), world gross revenues (r = 0,209, p < 0,05), domestic revenues (r = 0,182, p < 0,05), foreign revenues (r = 0,220, p < 0,05), DVD sales (r = 0,253, p < 0,005), and Blue ray sales (r = 0,249, p < 0,05).

Reliability testing of the above-mentioned variables shows a cronbach’s alpha of 0.805, which indicates strong internal reliability. The normality of the variables was tested and shows a non-normal distribution for production budget (skewness 1,637 and kurtosis 2.299), domestic box office revenues (skewness 2,142 and kurtosis 5,042), foreign box office revenues (skewness 2,547 and kurtosis 6,879), world gross revenues (skewness 2,420 and kurtosis 6,127), DVD after sales (skewness 3,040 and kurtosis 11,197), Blue ray after sales (skewness 2.345 and kurtosis 5.479), and sentiment score (skewness 0,306 and kurtosis -0,808).

The variables do not meet the assumption of normality; however, they do meet the central limitation theorem, which states that as the sample gets larger the variables can be assumed to be normally distributed (Doane & Seward, 2005, p. 299) The mediating effect of sentiment scores in the relationship between production budgets and world gross/after sales is tested using the Baron & Kenny (1986) method.

(34)

6. Results

In this section, the results from experiments run on the previously mentioned

variables are described. The hypotheses, formulated in chapter 3 (Literature review), are tested using the Barron & Kenny (1986) method described in the previous chapter (4) Research design and methods.

The impact of sentiment scores on box-‐office revenues

First, the direct effect of production costs was tested on world gross revenues, however sequel status is used as a control variable. The model summary shows an explanatory power of 66,4%, the adjusted R2 is 66,2%, and a durbin-watson statistic of 2,015, which both are acceptable. There is a significant relationship (F = 289,306, p < 0,05), between the two variables. The standardized beta coefficient of production budget shows an effect of 0,622. Thus, for every dollar spent on the production

budget, world gross revenues will grow by $ 0,622. The control variable sequel has an effect of 0,277. Therefore, for every dollar spent on a sequel, world gross revenues will rise with $ 0,277. Collinearity statistics for production budget (VIF = 1,197) and sequal (VIF = 1,197) show that the relationship does not suffer multi-collinearity.

Second, the relationship between sentiment scores and production budget shows an explanatory power of 16,9%, an adjusted of R2 16,3%, and a durbin-watson statistic of 2,140. Both acceptable. Similar to the relationship between production costs and world gross revenues, there is a significant relationship between sentiment score and production budget (F = 29,717, p < 0,05). Sentiment score has a

(35)

world gross revenue will grow by $ 0,065. Control variable sequel has an effect of 0,396. In other words, for every dollar spent on the sequel, world gross revenues will grow with $ 0,396. Collinearity statistics for sentiment score (VIF = 1,022) and sequel (VIF = 1,022), prove that muliti-collinearity is not a problem.

Third, the effect of sentiment score on world gross revenues has an

explanatory power of 31,4%, the adjusted R2 shows a model fit of 31%, and a durbin-watson statistic of 1,919 (acceptable). This effect is statistically significant (F = 67,197, p < 0,05). The sentiment score has a standardized beta effect of 0,132 on world gross revenues. Thus, for every dollar spent towards improvement of sentiment score, world gross revenue will grow with $ 0,132. The control variable sequel has an effect of 0,526. Thus, for every dollar spent on a sequel, the world gross revenue grows with $ 0,526. Collinearity statistics for sentiment score (VIF = 1,022) and sequel (VIF = 1,022) do not indicate multi-collinearity.

Fourth, the effect of all variables was tested. The model shows 67,2

explanatory power, an adjusted R2 of 66,8%, and a durbin-watson statistic of 2,001 (acceptable). The effect is statistically significant (F = 199,135, p < 0,05). For every dollar spent on the production budget, the world gross revenues will grow with $ 0,656. Every dollar spent on sentiment score improvements will generate $ 0,090. Control variable sequel has an effect of 0,266, thus every dollar spent on a sequel will result in a $ 0,266 increase in world gross revenues.

In summary, sentiment scores have a significant mediating effect on the relationship between production budget and box office gross revenues. Therefore, hypothesis one is rejected. Next, the mediating effect of sentiment scores on the relationship between production costs and after sales is tested.

(36)

The impact of sentiment scores on after sales revenues

First, the direct relationship between the variables production budget and after theatre DVD sales was tested. The model has 30,7% predictive power, with an adjusted R2 of 30,2%, and a durbin-watson statistic of 2,138 (acceptable).The effect is statistically significant (F= 53,241, p < 0,05). Every dollar spent on production budgets will increase DVD sales by $ 0,383. Each dollar spent on a sequel will generate $ 0,281 revenue in DVD sale .The collinearity statistics indicates no presence of multi-collinearity in either variable (VIF = 1,172).

Second, the effect of the variable sentiment score on production costs was tested. According to the results, the model has a predictive power of 16,9%, an

adjusted R2 of 16,3%, and an acceptable durbin-watson statistic of 2,140 (acceptable). This effect is statistically significant (F = 29,717, p < 0,05). Standardized coefficient beta’s show that every dollar spent on sentiment score will generate $ 0,065. For each dollar spent on a sequel, $ 0,396 is generated. Multi-collinearity is disregarded since the VIF for both variables is 1,022.

Third, the direct effect of sentiment score on DVD sales was tested. The predictive power of the model is 21,1%, the adjusted R2 20,4%, and the durbin-watson 2.045 (acceptable). Statistical significance is established (F = 32,028, p < 0,05). Sentiment score will generate $ 0,172 of DVD sales for every dollar spent. A dollar spent on a sequel will increase DVD sales revenues with $ 0,391. The VIF of 1,045 indicates that there are no problems with multi-collinearity.

Fourth, the mediating effect of sentiment scores on the relationship between the variables production budget and DVD sales was tested. Results show a prediction value of 32,4%, an adjusted R2_{of 31,5%, and a durbin-watson statistic of 2,144} (acceptable). The effect is statistically significant (F = 38,099, P < 0,05). The

(37)

standardized coefficient shows that for every dollar spent on sentiment score improvement, after DVD sales revenues will increase by $ 0,131. Control variable sequel has an incremental effect of $ 0,259. Collinearity statistics for sentiment score (VIF = 1,060) and sequel (VIF = 1,200) excludes multi-collinearity problems.

In summary, similar to hypothesis one, sentiment scores have a significant mediating effect on the relationship of production budgets and after theatre DVD sales. These results indicate that sentiment scores should be taken into consideration when distribution studios want to increase box-office and after sales revenues.

(38)

7. Discussion and managerial implications

Discussion

The sentiment scores of movies have a statistically significant mediating effect on the production budget and revenue relationship. These results indicate that the

improvement of sentiment scores on micro-blogs can be utilized to ultimately increase revenue streams. According to Jansen et al. (2009), brand management through

micro-blogging should be integrated with the overall marketing strategy and corporate branding. They go on to argue that micro-blogs can be used as information-pushing channels, in order to inform consumers about products or services. However, they caution that this tactic will only reach “the core brand audience”. Another effective way of utilising micro-blogs suggested is to use tweets as a customer feedback source. Furthermore, tweets allow for direct consumer interaction and a way to understand consumer preferences (p. 8).

When posting tweets, people can have countless intentions. The effect of posting from two possible starting points of tweets is an interesting area for future research. First, start on a hashtag-level; the post is intended to contribute to a discussion that already excites; identified by Pak & Paroubek (2010) as a public statement. The subject of the discussion is clear. Second, start from comment-level; the tweet is intended to comment on an event that is not yet an established topic. The subject of the discussion is still unclear and the tweet functions as an establisher of meaning; this type is categorized as a personal message by Pak & Paroubek (2010). Understanding the relationship between the tweet starting points, and the results presented during the study will contribute to decision making in regards to the establishment of franchises and projected success of movies.

(39)

The present study has several limitations. First, hashtags are not only used to tweet about the movie, but also for tweets about topics other than the movie. For example “@susanpenhaligon hey Susan, you never mentioned you high class soap supplies before! x http://t.co/dKMFpVfmRS”. This tweet was hashtagged with a movie title, however the movie is not mentioned. 1,500 tweets were loaded for each movie, which means that tweets similar to the example can lead to tainted results. Second, acrimony was not used during analysis, which means the analysis did not take sarcasm into consideration during the scoring of tweets. Future research needs to establish the effect of sarcasm on sentiment scores and how this affects the mediating effect on production budgets and revenues, in order to create more accurate predictive tools. Third, emoticons that could not be translated were removed. Consumers use emoticons to help others understand the unspoken meaning of their message. This means that emoticons not taken into consideration strips a layer of sentiment from the analyses. Future research should have a more extensive list of emoticon descriptions, so a larger set of emoticons are taken into consideration during emoticon translation. Fourth, re-tweets were not removed from the sample. Re-tweeting “is the process of copying another user’s tweets and posting to another account”. This means that extra weight might have been given to particular tweets (Go, Bhayani, & Huang, 2009, p.4). Future studies should investigate whether there is a significant difference in the impact of sentiment scores on revenues when re-tweets are not taken into

consideration.

Managerial implications

Currently the accepted usage of hashtags is to match the hashtag with the movie title. Which means that movies with generic names result in generic hashtags. These

(40)

movies have a multitude of tweets which do not apply to the movie itself. Adding the word “movie” to the hashtag search partially solved this problem. However, movies with a unique name, such as the addition of 2 (to indicate the sequel number) or 3D result in unique hashtags. Another implication of generic movie titles is the fact that the tweets could also apply to movies released in another year.

Tweets are generating and stimulating movements all over the world, allowing people interested in the same subject to easily follow updates. The sentiment of tweets published on the release date of a movie could help predict revenues. However, Jansen et al. (2009) caution organizations and warn that the micro-blogging branding is fluid, which results in the need for “constant and continual management”. This fluidity is a critical difference micro-blogging has compared to “other forms of product expression” (p. 9).

(41)

8. Conclusion

The present study was focused on the research question “how can movie distribution studios effectively utilize buzz and word-of-mouth in order to increase box-office and after theatre revenues of released movies?” Consumers prefer organic WOM over amplified WOM, since the power of WOM is based on the credibility of consumer to consumer references. Social media platforms create an opportunity for movie

distribution studios to allow consumers to review and share movies they have seen with their friends, family and acquaintances. This not only creates an opportunity to lower marketing costs, it also allows the news of the movie release to travel further into the social network of interested and aware consumers, since consumers who are interested in the movie probably have similar others clustered in their social media networks, which creates the opportunity for macro WOM to emerge.

Results show that (1) sentiment scores have a significant mediating effect on the relationship between production budget and box office gross revenues and (2) sentiment scores have a significant mediating effect on the relationship of production budgets and after theatre DVD sales. This indicates that generating eWOM through micro-blogs focused on increasing sentiment scores can ultimately help increase revenue streams.

The present study has the following limitations. First, sentiment on movies was established through tweets with hashtags related to the movie title. However, hashtags are not only used to tweet about the movie, but also for tweets about topics other than the movie. Second, acrimony was not considered during analysis, which means the analysis did not take sarcasm in consideration during the scoring of tweets. Third, emoticons that could not be translated were removed, which means a layer of

(42)

sentiment was not considered. Fourth, re-tweets were not removed from the sample, which may have resulted in extra weight being given to such tweets.

(43)

9. References

Agarwal, A., Xie, B., Vovsha, I., Rambow, O., & Passonneau, R. (2011, June). Sentiment analysis of twitter data. In Proceedings of the Workshop on Languages in Social Media (pp. 30-38). Association for Computational Linguistics.

Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of personality and social psychology, 51(6), 1173.

Basuroy, S., & Chatterjee, S. (2008). Fast and frequent: Investigating box office revenues of motion picture sequels. Journal of Business Research, 61(7), 798-803.

Basuroy, S., Desai, K. K., & Talukdar, D. (2006). An empirical investigation of signaling in the motion picture industry. Journal of Marketing Research, 287-295.

Breen, J (2011). Things I tend to forget. Retrieved 24-5-2015, from:

https://jeffreybreen.wordpress.com/2011/07/04/twitter-text-mining-r-slides/

Brown, J. J., & Reingen, P. H. (1987). Social ties and word-of-mouth referral behavior. Journal of Consumer Research, 350-362.

Cunha, E., Magno, G., Comarela, G., Almeida, V., Gonçalves, M. A., & Benevenuto, F. (2011, June). Analyzing the dynamic evolution of hashtags on twitter: a

(44)

language-based approach. In Proceedings of the Workshop on Languages in Social Media (pp. 58-65). Association for Computational Linguistics.

Daugherty, T., & Hoffman, E. (2014). eWOM and the importance of capturing consumer attention within social media. Journal of Marketing Communications,20(1-2), 82-102.

Dhar, T., Sun, G., & Weinberg, C. B. (2012). The long-term box office performance of sequel movies. Marketing Letters, 23(1), 13-29.

Doane, D. P., & Seward, L. E. (2005). Applied statistics in business and economics. USA: Irwin, 299 - 303.

Duan, W., Gu, B., & Whinston, A. B. (2008). The dynamics of online word-of-mouth and product sales—An empirical investigation of the movie industry.Journal of retailing, 84(2), 233-242.

eBIZ MBA (2014, February). Top 15 most popular social media network sites February 2014. Ebizmba.com. Retrieved 6-2-2014, from

http://www.ebizmba.com/articles/social-networking-websites

Elberse, A., & Eliashberg, J. (2003). Demand and supply dynamics for sequentially released products in international markets: The case of motion pictures. Marketing Science, 22(3), 329-354.

(45)

Eliashberg, J., & Shugan, S. M. (1997). Film critics: Influencers or predictors?.Journal of Marketing, 61(2)

Ferrara, E., De Meo, P., Fiumara, G., & Provetti, A. (2012). The role of strong and weak ties in Facebook: a community structure perspective. arXiv preprint

arXiv:1203.0535.

Ferrara, E., De Meo, P., Fiumara, G., & Provetti, A. (2012). The role of strong and weak ties in Facebook: a community structure perspective. Preprint at< http://arXiv. org/abs/1203.0535.

Github (2015) Retrieved 9-6-2105 from: https://github.com/jeffreybreen/twitter-sentiment-analysis-tutorial-201107/blob/master/R/sentiment.R

Gladwell, M. (2000). The science of the sleeper. BOOKSELLER, 22-25.

Go, A., Bhayani, R., & Huang, L. (2009). Twitter sentiment classification using distant supervision. CS224N Project Report, Stanford, 1, 12.

Granovetter, M. S. (1973). The strength of weak ties. American journal of sociology, 1360-1380.

Granovetter, M. S. (1973). The strength of weak ties. American journal of sociology, 1360-1380.

(46)

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.

Holbrook, M. B., & Addis, M. (2007). Taste versus the market: An extension of research on the consumption of popular culture. Journal of Consumer Research, 34(3), 415-424.

Jansen, B. J., Zhang, M., Sobel, K., & Chowdury, A. (2009, April). Micro-blogging as online word of mouth branding. In CHI'09 Extended Abstracts on Human Factors in Computing Systems (pp. 3859-3864). ACM.

Karniouchina, E. V. (2011). Impact of star and movie buzz on motion picture

distribution and box office revenue. International Journal of Research in Marketing, 28(1), 62-74.

Kimmel, A. J., & Kitchen, P. J. (2014). WOM and social media: Presaging future directions for research and practice. Journal of Marketing Communications, 20(1-2), 5-20.

Kouloumpis, E., Wilson, T., & Moore, J. (2011). Twitter sentiment analysis: The good the bad and the omg!. ICWSM, 11, 538-541.

(47)

Krider, R. E., & Weinberg, C. B. (1998). Competitive dynamics and the introduction of new products: The motion picture timing game. Journal of Marketing Research, 1-15.

Kuijken, B., Leenders, M. A., Wijnberg, N. M., & Gemser, G. (2013, January). How musical festival success is affected by differences in consumer and producer

classification. In Academy of Management Proceedings (Vol. 2013, No. 1, p. 14382). Academy of Management.

Libai, B., Bolton, R., Bügel, M. S., De Ruyter, K., Götz, O., Risselada, H., & Stephen, A. T. (2010). Customer-to-customer interactions: broadening the scope of word of mouth research. Journal of Service Research, 13(3), 267-282.

Liu, Y. (2006). Word of mouth for movies: Its dynamics and impact on box office revenue. Journal of marketing, 74-89.

Mohr, I. (2007). Buzz marketing for movies. Business Horizons, 50(5), 395-403. Mol, J. M., & Wijnberg, N. M. (2007). Competition, selection and rock and roll: The economics of payola and authenticity. Journal of Economic Issues, 701-714.

Nasukawa, T., & Yi, J. (2003, October). Sentiment analysis: Capturing favorability using natural language processing. In Proceedings of the 2nd international conference on Knowledge capture (pp. 70-77). ACM.