I HEARD IT THROUGH THE DIGITAL
GRAPEVINE: AN EMPIRICAL
INVESTIGATION OF THE EFFECTS OF
ELECTRONIC WORD-OF-MOUTH ON
SALES IN THE FILM INDUSTRY
Master’s Thesis Bart Penris | 6178545
MSc Business Administration | Entrepreneurship & Management in the Creative Industries Supervised by: Ms. I. Rozenthale
Second reader: dr. B. Kuijken June 24, 2016
2
STATEMENT OF ORIGINALITY
This document is written by Student Bart Penris 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
In the past five months a great deal of my time went to this thesis. It is by far the most extensive, complicated, challenging but also the most satisfying project I have done during my years at the University of Amsterdam. Completing this project therefore truly feels like a personal milestone. I do however realize that it would not have been possible without the help, support and understanding of several people. I would like to thank a few of them.
My supervisor, Ms. Ieva Rozentale, for her advice, creative thinking and useful feedback throughout the whole project.
Mattijs Grannetia from Boxofficenl.net, for providing me with the full 2014 – 2015 FDN dataset.
Jaco van de Pol, Emile Pater and Leon Horbach for their participation in the coding process.
My girlfriend for her overall support and the many days we spend together in the library.
My friends for their advice and encouragements.
My brothers for their support and understanding.
4
Abstract
Consumers who participate in discussions and post their opinions on products online engage in electronic word-of-mouth (eWOM) communications. eWOM is believed to be an important factor in explaining sales of experience goods because it is perceived as an independent source of information free of corporate interests. The aim of this study was to analyze how the volume and valence of eWOM from multiple social media sources affected the sales of
experience goods in the setting of the Dutch film industry. Furthermore, because earlier research indicated that the effects might be different depending on the type of film, we made the distinction between the categories mainstream, niche and unclear. We used a sample of 363 films to study these effects with multiple regression analyses. We checked for differences in the pre-release and cumulative period to better isolate the effects. Our findings suggest that volume is the most important eWOM factor in explaining sales in both time periods and for all film types, while positive valence only matters in the cumulative period. Contrary to what we expected, we found no important differences for the effectiveness of eWOM volume between mainstream and niche film types. Positive eWOM valence appears to be only significant for mainstream and unclear film types, while critic ratings only matter for niche films. Interestingly enough we also found that negative eWOM valence can sometimes be beneficial in the pre-release period, especially for films that do not classify as a typical mainstream or niche offering. This study finds further evidence for the awareness effect of eWOM as found in earlier research and especially contributes in the sense that it used a social media monitoring tool to collect eWOM data on a very large scale from various sources. Simultaneously, our findings and implications indicate that further research is needed to gain additional insight in the dynamics between eWOM, experience goods and commercial performance.
5
Table of Contents
List of tables ... 7 List of figures ... 7 1. Introduction ... 8 2. Literature review... 102.1 Creative goods as experience goods ... 11
2.2 Critics: a traditional third party ... 12
2.3 eWOM ... 14
2.4 eWOM on social media ... 16
2.5 Volume and valence of eWOM ... 18
Volume ... 19
Valence ... 20
2.6 Mainstream versus niche ... 22
3. Research design and Methodology ... 25
3.1 Research design ... 26 3.2 Data collection ... 27 3.3 Dependent variables ... 30 3.4 Independent variables ... 30 3.5 Moderating variable ... 31 3.6 Control variables ... 32 3.7 Method of analysis ... 33 4. Results ... 36 4.1 Descriptive statistics ... 36 4.2 Correlations ... 39 4.3 Hierarchical regressions ... 40 Main effects ... 40
6
Interaction effects ... 43
4.4 Split file regression ... 44
5. Discussion ... 47
5.1 Findings and theoretical foundations ... 48
Volume ... 48
Valence ... 49
Film type ... 50
5.2 Limitations, suggestions for future research and contributions ... 53
5.3 Practical implications ... 56
6. Conclusion ... 57
References ... 59
Appendices ... 65
Appendix A: List of films used in final sample ... 65
Appendix B: Instructions for Coders ... 72
7
List of tables
Table 1: Descriptive statistics & ANOVA………....…37
Table 2: Post-Hoc tests.………..…..38
Table 3: Correlations…….………40
Table 4: Regression analysis – Opening Weekend B.O.………...…………..…….42
Table 5: Regression analysis – Total B.O.………43
Table 6: Regression analysis – Split file………...47
Table 7: Summary of results………...………...…48
List of figures
Figure 1: Conceptual Model……….…258
1. Introduction
“All motion pictures are a gamble. Anything having to do with creating something that nobody’s seen before, and showing it, and counting on 10 or 20 million people, individuals, to go into the theater to make or break that film... that’s a gamble” (cited in: Lampel &
Shamsie, 2000, p. 239).
This quote by film director Steven Spielberg demonstrates the high uncertainty that surrounds both producers and consumers of creative goods. Creative goods like films are extra sensitive to such uncertainties because they can be classified as experience goods. While their counterparts, search goods, have certain characteristics that make them mutually comparable, the quality and value of an experience good can only be determined after buying and experiencing it (e.g. Cui, Lui & Guo, 2012). Thus, consumers look for alternative methods to assess the quality of an experience good before a purchase. A traditional way of doing this is by reading the reviews of professional critics in newspapers or magazines. Creative industries like art, theater and film all have dedicated critics who evaluate and advise the public on new releases and by doing so act as selectors in their industry (Wijnberg, 1995). Such critics are seen as third parties because they are supposed to be an independent source of information for consumers (Chen, Liu & Zhang, 2012).
Apart from seeking information from professional critics, consumers also mutually advise each other on different kinds of products. This process is called word-of-mouth (WOM; e.g. Day, 1971). The rise of the Internet increased the speed through which reviews and opinions spread across people from all over the world and simultaneously opened a whole new virtual platform for consumers to share their own opinions and experiences about products. These technological developments ensured that part of the discussions about products nowadays takes place online. This has led to a new term being used in academic literature: electronic word-of-mouth (eWOM). eWOM is similar to regular WOM, the difference being that eWOM happens
9
online, for instance through blogs, forums or review websites (e.g. Liu, 2006; Hennig‐Thurau, Gwinner, Walsh, & Gremler, 2004).
The rise of social media in particular has accelerated and extended these dynamics. Social media have made it even easier for consumers to post opinions and reviews online themselves and by doing so act as selectors for their own social environment. Consumers who engage in such activities should not solely be seen as consumers, but also as an external third party that deserve the attention of managers. While regular websites already offered these possibilities, they often require an additional registration before people are allowed to post or discuss anything. Social media have lowered the threshold to participate in this process by allowing and even encouraging real-time sharing of content and text on any given subject, without requiring an additional registration. This may lead to the expression of opinions and statements that are closer to people’s real-life feelings and beliefs (Hennig-Thurau, Wiertz & Feldhaus, 2014).
Scholars have been exploring the links between person-to-person communications and business outcomes in many different settings. Both WOM and eWOM effects are believed to be of great importance for the sales of experience goods (e.g. Liu, 2006; Chevalier & Mayzin, 2006). However, in almost every study on this matter the actual eWOM data is collected from a single source, such as a dedicated forum or website. To our knowledge there are very little studies on the impact of eWOM on business outcomes in which the eWOM data is derived from multiple social media sources. Because social media eWOM may have different implications for business outcomes, this opens up an interesting gap for a study, which is formulated in the following research question: how does multiple source eWOM relate to the box office sales of films?
Earlier studies suggest that the extent to which a third party affects the success of a creative good is also partly dependent on the type of meta category the good belongs to (e.g.
10
Gemser, Oostrum & Leenders, 2007). Creative goods can be typically divided into two ‘meta’ categories: one consisting of popular - or mainstream - products, and the other one containing more niche products (Brynjolfsson, Hu & Simester., 2011). To our knowledge this implication has only been studied in the context of professional critics, and not yet with multiple source eWOM. This leads to the second research question: how is the effect as described in the first research question different for mainstream and niche films?
The setting of this study is the Dutch motion picture industry. This industry is particularly interesting for a study on eWOM because films are frequently discussed among consumers and highly sensitive to word-of-mouth interactions (Eliashberg, Jonker, Sawhney & Wierenga, 2000). We use a sample of 363 films released during a two-year period in cinemas all across the country. We distinguish between the pre-release and cumulative time period, both on the side of the dependent variable (box office) as the independent variable (eWOM), in order to isolate the effects of eWOM on sales in two more specific timeframes. Next to the theoretical contributions, this study also informs managers in the creative industries on the implications of multiple source eWOM on the commercial performance of different types of creative goods. Therefore, it can help with the development of digital marketing strategies in creative industries. This thesis is structured as follows: first, an overview of the literature on creative goods and eWOM will be given. Second, the specific research design, data collection and methodology of this study are stated. Third, the results and findings are presented and lastly, the limitations and managerial implications of this study are covered, as well as suggestions for future research.
2. Literature review
In this chapter previous academic research on the topics covered in this thesis will be explored. First, the specific characteristics of creative goods and the role of both traditional and emerging
11
third party influencers will be elaborated on. Second, we will move to eWOM and specifically eWOM on social media and the different effects that volume and valence are believed to have on sales. Lastly, the distinction between mainstream and niche in the context of the film industry will be explored.
2.1 Creative goods as experience goods
Creative goods like films traditionally classify as experience goods. These goods have to be experienced (and thus, bought) in order to determine the quality. Furthermore, even after buying an experience good, any value judgement will be subjected to personal preferences of the consumer (Cui et al., 2012). As Caves (2003, pp. 74) states: “Producers make many decisions that affect the expected quality and appeal of the product, yet their ability to predict its audience's perception of quality is minimal”. This is called the ‘nobody knows’ principle, which simply means that nobody really knows the true value of a creative good. (Caves, 2003). Furthermore, because consumer preferences are unstable, production decisions depend on educated guesswork and prior success of a good does not provide reliable guidelines for new products (Lampel & Shamsie, 2000). This leads to a high uncertainty for producers and entrepreneurs in these industries, because the likeability and success of the products they release will affect both their artistic reputation and financial situation.
At the same time consumers of experience goods face uncertainties as well. When consumers for example go to a cinema, they pay upfront without knowing if they are going to like the product. If they do not like it, there is no chance of returning the product and getting a refund. Therefore, potential consumers search for ways to assess the value and quality of an experience good before a purchase in order to reduce their uncertainty. Towse (2011, pp. 213-214) notes that there are three levels on which this uncertainty reduction happens. First, there are advertising messages in which producers make claims on the features of their goods.
12
However, advertising is often treated with skepticism and sometimes even leads to a higher uncertainty because these claims can still not be verified without purchasing the product (Lampel & Shamsie, 2000). Second, consumers sometimes can get limited access to experience goods in the form of trials and previews. Listening to snippets of new music and playing demos of videogames are examples of this, but this is not possible for every product. Third, consumers use publicly stated preferences of others as predictors of their own experience.
This last method is crucial because it is often the only available independent source of information on experience goods, especially when the option of a preview or trial is not available, like is the case with films. The individuals or groups who publicly state their opinions and preferences about experience goods can be seen as surrogate consumers who take on the task of reviewing products that may be of interest for the target group they reach out to (Hirsch, 1972). This means that the value of experience goods is often subjectively measured through a set of preferences and standards of these groups or individuals that act as selectors in the market (Wijnberg & Gemser, 2000). They form third parties who communicate their opinions about experience goods to potential consumers. The assumption that such third parties are often the only independent source of information available puts them in a position of power, and in fact “they become participants in the strategic rivalry that shapes the industry” (Lampel & Shamsie, 2000, p. 238).
2.2 Critics: a traditional third party
Traditionally, the only way for potential consumers to get an indication of the quality of an experience good was through the printed reviews of professional critics. In such reviews, critics talk about the background of the concerned good, the content and typically end with a value judgement (Brown, Camerer & Lovallo, 2012). The reviews are published in newspapers, magazines and online and often appear just before the goods enters the market. In order to have
13
the review published prior to the release, critics are sometimes given the chance to get exclusive access before regular consumers.
Because of the timing of the reviews and the value judgement they contain it is no surprise that scholars in the past hypothesized that critics have an effect on the success of the product they review. The specific effects and directions seem to be partially dependent on the type of product and the industry. Friberg and Grönqvist (2012) found a positive effect of favorable reviews of a wine on the sales of that particular wine. A study by Heiman (1997) shows a similar result for restaurants. For books however, research indicates that mutual disagreement of critics and the expression of extreme judgements, positive or negative, has a positive effect on sales (Clement, Proppe & Rott, 2007). A more recent study on reviews in the book industry suggests that negative reviews can even enhance sales when prior awareness about the book is low (Berger, Sorensen & Rasmussen, 2010).
The effects of critic reviews have also been studied in the context of the film industry. Litman (1983) finds that critic rating is one of the three most important determinants of cumulative box office, next to the use of a major distributor and a Christmas release. Wallace, Seigerman and Holbrook (1993) also conclude that critics influence sales, and claim that the relationship is U-shaped. This means that for positively reviewed films higher ratings lead to more revenues, but also that films with the most negative ratings earn more revenues than films with mixed reviews. Their finding partially conflicts with the study of Wanderer (1970), who argues that the tastes of critics are similar to that of audiences. However, there has also been research that suggests positive reviews have no significant influence at all (Ravid, 1999).
A crucial question that is addressed in multiple papers on critics in the film industry is whether the possible effect that their reviews have on box office sales should be interpreted as a prediction effect or as an influencer effect (Eliashberg & Shugan, 1997; Gemser et al., 2007; Basuroy, Chatterjee & Ravid, 2003; Reinstein & Snyder, 2005). The difference between the
14
two is that critics as predictors only predict the success in terms of box office of a to-be-released film, while critics as influencers actually influence people to go and see the film. Both potential effects are isolated by looking at the effects of reviews on opening-weekend box office (influencer effect) and cumulative box office (prediction effect).
Eliashberg and Shugan’s (1997) findings suggest the existence of a prediction effect but they find no evidence for an influencer effect. Interestingly, this is somewhat contradictory to the results of Basuory et al. (2003), as they find evidence for both the prediction and the influencer effect. Reinstein and Snyder (2005) also find both effects to be significant, but agree with Eliashberg and Shugan that the prediction effect is more important. They also add that a higher perceived expertise of the critic can strengthen both effects, and that the significance of the effects is dependent on the type of film. This last finding is confirmed by Gemser et al. (2007) and its implications will be discussed in section 2.6.
It appears that scholars in the past had mixed conclusions about the influence of critics on the sales of experience goods. In this study the focus will not be on critics and their role as selectors and third party influencers of the public. However, the above listed findings may still be of great importance because they do indicate the importance of third parties for experience goods and can help to hypothesize the effects of other parties than professional critics on sales in the creative industries.
2.3 eWOM
Now that the majority of the western world has access to the internet, the reviews of
professional critics are widely available to consumers. A much more radical result however is that consumers now have the chance to also write reviews and statements in a
non-professional manner. The mutual communications of consumers about products online is called electronic word-of-mouth (eWOM).
15
The roots of eWOM can be found in the concept of word-of-mouth (WOM). We speak of WOM when there is oral, person-to-person communication regarding a brand, product or service between a communicator and a receiver (Amdt, 1967). Crucial in this matter is that the receiver perceives the communicator as non-commercial, because this is how WOM
distinguishes itself from advertising. Generally, WOM is believed to be of great importance for sales. Sheth (1971) even claims that when the objective is to raise awareness for an innovation or to persuade the public in trying a product, WOM is more important than
advertising and critics. WOM is believed to be so effective because the receiver perceives the source of information as credible and trustworthy (Day, 1971). Bone (1995) adds that WOM is free from corporate interests and marketing intentions. Because experience goods are often consumed collectively and feature in daily conversations, WOM is believed to be of even greater importance in creative industries (Eliashberg, Elberse & Leenders, 2006).
The rise of the internet has created a new dimension in person-to-person communications in a virtual world. This new, some may say modernized way of WOM is called eWOM. It can be defined as “any positive or negative statements made by potential, actual, or former customers about a product or company, which is made available to a multitude of people and institutions via the Internet” (Henning-Thurau et al., 2004, pp. 39). These statements are typically ventured through blogs, forums or review sites, but can also take on other forms. eWOM builds on the same basis as WOM, while the fact that it takes place online brings some additional implications. The online character of eWOM allows people who engage in these discussions to maintain their anonymity (Lee & Youn, 2009). This leads to a situation where consumers are more comfortable with expressing their true opinions and experiences (Goldsmith & Horowitz, 2006).
The subjective nature of experience goods suggests that eWOM still does not have the ability to objectively inform consumers on the features and quality of such goods. However,
16
eWOM can raise the expectations and create a certain buzz around a product. When looking at the film industry in particular, studios sometimes even heavily count on positive WOM / eWOM in their marketing strategies for films. Some films are released on a relatively low number of screens and the studio plans to gradually expand the number of screens following the positive eWOM a film gets. This release strategy is called a platform release (e.g. Reinstein & Snyder, 2005). In this case eWOM is also part of the marketing mix by feeding discussions and generating awareness at the target group. The specific implications of eWOM on business outcomes will be discussed more extensively in section 2.5.
Chu and Kim (2011) argue that the eWOM behavior of consumers has three building blocks: opinion seeking, opinion giving and opinion passing. The first group contains consumers who search on the internet for information and opinions before they make a purchasing decision. The second group represents the traditional concept of an opinion leader. As discussed earlier, this can be a professional critic, but also regular consumers who are for example active on product review sites or forums. The third group of opinion passing consumers are especially important for eWOM, because opinion passing behavior leads to the wide availability of opinions and corresponding discussions on the internet (Dellarocas, 2003).
2.4 eWOM on social media
Social media channels like Facebook, Twitter and Instagram provide individuals with new opportunities to engage in discussions with their peers about products, brands or services on the internet. They are designed to spread and share content with the users’ network within seconds. Therefore, they are believed to facilitate and further enhance the process of opinion seeking, opinion giving, and especially opinion passing behavior that form eWOM (Chu & Kim, 2011). Peer-to-peer communications through social media also have the advantage that the sources of information in this case already belong to or can be identified through someone’s network (Chu
17
& Kim, 2011). While this diminishes the potential advantage of anonymity in eWOM communications, it enhances the perceived credibility and trustworthiness of the communicator (Chu & Kim, 2011; Kim, 2014).
A crucial difference between ‘regular’ eWOM and eWOM on social media lies in the motivations for users to participate and engage in discussions. The channels of regular eWOM such as dedicated websites and forums are likely to be used by consumers who are actively seeking information, advice or opinions on a specific product. This is confirmed by Hennig-Thurau et al. (2004), who find that advice seeking, positive self-enhancement, concern for other customers and economic incentives are all positively related to both the visit frequency and the number of comments written on a web-based opinion platform. Goldsmith and Horowitz (2006) also find that financial risk reduction factors are important factors for engaging in online discussions, and thereby confirm the suggestion that participants in ‘traditional’ eWOM often have specific goals to pursuit.
The motivations for social media usage are more casual than for regular eWOM because for many people social media are part of their everyday life (Kim, 2014). Therefore, the goal of social media usage is not to find opinions on specific products like is the case with a regular eWOM source. Furthermore, while forums and dedicated review websites often require an additional registration or membership before one can actively and fully participate in discussions, social media are already used by a majority of the western world, which means that all of their users can freely share information and engage in discussions with their peers (Kim, 2014). Social media interactions are therefore closer related to offline WOM. Hennig-Thurau et al. (2014) describe eWOM on social media almost as a lovechild of eWOM and WOM. They claim that it has the personal connection between sender and receiver, real-time transmission, continuous feedback and push and pull options characteristics of offline WOM, while the fact that it entails written communication and the receiver is a potentially very large group relates
18
to eWOM. Furthermore, they argue that social media messages are perceived by receivers as more honest and without an agenda than the traditional, more closed eWOM sources. Kim (2014) uses the same line of reasoning by suggesting that eWOM on social media has the credibility and trustworthiness advantages of offline WOM combined with the easy availability of eWOM.
Because of these reasons, eWOM on social media better represents customers’ perspectives on experience goods and potential consumers will use this as a source while selecting and seeking products (Kim, 2014). Studying eWOM implications through multiple social media sources is therefore more likely to give a comprehensive representation of how people feel and act and it is more effective when trying to capture the behavior of the human being in its natural habitat.
2.5 Volume and valence of eWOM
eWOM can be interpreted and measured in multiple ways. Two characteristics of eWOM that have been used a lot in research are volume and valence. The volume of eWOM indicates the attention a product receives online and can be expressed in the number of messages (Hennig-Thurau et al., 2014) or quantity of information available for potential buyers (Cheung & Thadani, 2012). The valence, or sentiment, of eWOM indicates whether the message contains a positive or a negative value judgement (e.g. Godes & Mayzlin, 2004; Chen et al., 2012; Cui et al., 2012). The volume and valence of eWOM are also the main concepts studied in this thesis. In this section, studies on the effects of volume and valence are reviewed and based on these earlier findings hypotheses are formulated.
19
Volume
Overall, the volume of eWOM is believed to be an important determinant of sales. Liu (2006) studied the effects of eWOM in the form of content on Yahoo Movies on the box office sales of films. He found that eWOM is most active in the week prior to the release of a film and that especially the volume of eWOM significantly relates to box office performance. Furthermore, he found that pre-release eWOM volume has significant explanatory power over opening weekend box office, and continues to do so for aggregated box office. This confirms the idea that eWOM volume can create a certain buzz which is beneficial to both initial and cumulative sales. Kim (2014) extends Liu’s research by taking multiple source eWOM into account. Her findings also suggest that the volume of eWOM positively relates to box office performance. Duan, Gu and Whinston (2008) speak of an ‘awareness effect’ created by the volume of online postings about films which significantly influences box office.
Research in the setting of other creative industries has also highlighted the importance of eWOM volume. Dhar and Chang (2009) show that the volume of eWOM in the form of blog posts is positively correlated with future music sales. Amblee and Bui (2011) found that for e-books, the volume of eWOM has the power to convey the brand and product reputation and influence sales. For the sales of regular books on the internet the volume of eWOM also shows to have a significant impact (Chevalier & Mayzlin, 2006). These findings are further confirmed by Cui et al. (2012), who find that for experience goods the volume of user reviews matter the most in predicting sales.
Because the volume of eWOM is believed to be an important predictor of cumulative sales but also has the power to create a buzz in the pre-release period, we can formulate the first two hypotheses:
20
H1a: The volume of eWOM in the pre-release period has a positive effect on opening weekend box office sales.
H1b: The total volume of eWOM has a positive effect on cumulative box office sales.
Valence
The impact of eWOM valence has also been studied in a wide range of industries and settings. Lee and Youn (2009) state that negative user reviews are harmful to the extent a consumer would be willing to recommend retail products to friends. Additionally, Chevalier and Mayzlin (2006) found that the negative impact of new one-star reviews on book sales is greater than the positive impact of new five-star reviews. Park and Lee (2009) claim that while negative eWOM damages experience goods, positive valence does no good. They argue that negative information magnifies consumers’ uncertainty and fear that derives from their poor cognitive knowledge about the product. These findings contribute to the notion of negativity bias, which simply suggests that people weigh negative information more than positive information (e.g. Rozin & Royzman, 2001). A possible explanation for the existence of the negativity bias is that positive reviews are more attributed to the personal preferences and characteristics of the reviewer, rather than the actual product experience (Chen & Lurie, 2013).
Other scholars have conflicting findings. Sonnier, McAlister and Rutz (2011) find that both positive valence (positively) and negative valence (negatively) of eWOM influence sales. However, their study is limited to search goods and therefore not useful for building our conceptual model. Cui et al. (2012) and Liu (2006) find that for experience goods the valence of eWOM does not matter at all, while Vermeulen and Seegers (2009) studied the effects of eWOM in the hotel industry and find that both positive and negative eWOM increases consumer awareness, especially for lesser known hotels. They even suggest that the positive eWOM does
21
more good than negative eWOM hurts, which directly conflicts with the concept of the negativity bias.
When looking at the particular case of eWOM in the film industry there is also no clear consensus on the impact of valence. Hennig-Thureau et al. (2014) only find a significant negative effect on sales for negative eWOM valence. Liu (2006) finds no significant effect between valence of eWOM and box office. Duan et al. (2008) have a similar conclusion. They do however suggest that although eWOM valence itself does not directly influence sales, there is an indirect effect on sales because more positive eWOM can lead to a higher eWOM volume. Dellarocas, Zhang and Awad (2007) show that positive eWOM valence is a significant predictor of future performance. Kim (2014) also finds a significant association between positive eWOM valence and box office sales. This idea is reinforced by Chintagunta, Gopinath and Venkataraman (2010), who state that positive eWOM valence is crucial in predicting box office sales.
Although the proposed impact of eWOM valence is not undisputed and sometimes even denied, it appears that studies where the methodological and theoretical approach is most congruent with ours tend to conclude that a higher level of positive eWOM leads to higher box office sales, in one way or another (Duan et al., 2008; Kim, 2014). Therefore, we expect similar results for positive eWOM valence as we do for eWOM volume. This expectation leads to the following two hypotheses:
H2a: Positive valence of eWOM in the pre-release period has a positive effect on opening weekend box office sales.
H2b: Positive total valence of eWOM in has a positive effect on cumulative box office sales.
22
2.6 Mainstream versus niche
We already elaborated on the major differences that separate search goods from experience goods. A typical further distinction that is made in many creative industries is the distinction between two ‘meta-categories’: mainstream and niche. Mainstream creative goods are expected to attract a large audience and gross high revenues, while niche creative goods are more specialized, and targeted at a smaller, often more intellectual audience (Brynjolfsson et al., 2011). By using this distinction, the results of the current study also become more generalizable to other creative industries than the film industry because the mainstream - niche categorization can be found in almost every one of them.
Before hypothesizing the different effect of eWOM for both types of films it is important to clarify the boundaries between the two. Scholars have used different approaches to do so in the past. Categorizations have been made based upon intrinsic features of the film itself, for example the content, genre or narrative structure (e.g. Bordwell & Thompson, 1997) or the presence of special effects and movie stars (e.g. Bagella & Bechetti, 1999). Other scholars have used extrinsic features, like the height of the production and marketing budget (Holbrook & Addis, 2008) or the size and identity of the distribution company responsible for distributing the film (Zuckerman & Kim, 2003).
Gemser et al. (2007) build upon the approach used by Zuckerman and Kim (2003). They determine whether a film should be considered mainstream or niche by looking at the type of cinema a film is being released in. Cinemas are used instead of distributors because Dutch film distribution companies often do not specialize in only one type of film (Gemser et al., 2007). Although at the time this may have been a reasonable approach, one must take into account that the Dutch cinema landscape has changed in the last years. The largest mainstream cinema operator that is currently active in the Netherlands, Pathé, introduced Pathé Alternative Cinema (PAC) in late 2005. The PAC program offers a more niche selection and responds to the
23
growing demand for quality films (Pathe, 2016). At the same time new cinemas have opened that don’t specifically focus on only one type of film, but rather offer both mainstream and niche films.
These recent developments in the Dutch cinematic landscape can be seen as a result of the emerging relevance of the concept of cultural convergence (e.g. Shrum, 1991). Advocates of this concept state that the once clear differences between high and popular art diminish due to the growth of the middle class in society. This makes it questionable to divide films into mainstream and niche on the basis of fixed factors like cinemas, studios or budgets. Therefore, in this study we chose to use coders to make the distinction between film types out of the belief that the best way a distinction between mainstream and niche films can be made nowadays is through the subjective opinions of regular consumers; the same subjective opinions that ultimately judge on the aesthetic value of those goods (Lampel & Shamsie, 2000).
To our knowledge, no studies have been done on the specific effects of social media eWOM volume and valence on mainstream and niche films. A study by Yang, Kim L., Amblee and Kim W. (2009) considers the effect of eWOM on mainstream and non-mainstream films in South Korea. While they find no significant difference for film type, their results do not form a good basis for this research because of the use of single source of eWOM in the form of a film-related website, in a similar way Liu (2006) and Duan et al. (2008) did earlier. The different film types have however been examined in the context of film critics. In the beginning of this chapter the prediction and influencer effects of such critics are discussed. Reinstein and Snyder (2005) show evidence for the existence of an influencer effect, especially for films with platform releases. Because of this limited release strategy, platform releases are often associated with niche films. Gemser et al. (2007) have a similar finding. They conclude that for mainstream films, critics have a prediction effect, while for niche films critics can actually influence the box office performance.
24
While none of these two studies uses eWOM as a predictor variable, its findings could still be of very good use for this study. Austin (1983) notes that the people who visit niche films tend to agree more with critics than the mainstream audience does. This could suggest that niche consumers also agree more with their fellow fans and therefore are more likely to be influenced by pre-release eWOM from each other. We therefore expect that the pre-release and post-release effects that are associated with the predictor and influencer effect of critics will hold for eWOM. This leads to the following two hypotheses:
H3a: For niche films, the effects of pre-release eWOM volume and valence will be significant for opening weekend box office, while the effects of total eWOM volume and valence on cumulative box office will be not significant.
H3b: For mainstream films, the effects of total eWOM volume and valence will be significant for total box office, while the effects of pre-release eWOM volume and valence on opening weekend box office will be not significant.
The different effect that eWOM potentially has on mainstream and niche films can be linked back to the long tail theory (e.g. Brynjolfsson et al., 2011). This theory states that because of the rise of the internet, it should become easier for the more obscure and niche products to gain attention and boost their sales. This would suggest that a higher level of especially eWOM volume would be extra beneficial for niche movies. Dellarocas and Narayan (2007) note that the volume of eWOM messages is higher for mainstream products because people tend to post more on these subjects because of the interplay with their peers. According to De Meyer (2012) this indicates that if producers of niche products would find a way to increase the eWOM volume on their products, this would be beneficial for their sales. This leads to the final hypothesis and the presentation of our conceptual model in figure 1:
25
H3c: The volume of eWOM will have a stronger effect on the box office of niche films, compared to mainstream films
Figure 1 Conceptual model
3. Research design and Methodology
In this section the research methods used to test the different hypotheses in this thesis are elaborated on. First, the overall research design and setting are discussed. Secondly, the data collection method, the used sample and the measurement of the variables are described. We conclude with presenting the specific methods of statistical analysis that were used to test the hypotheses.
eWOM
Volume
Valence
Box office sales
26
3.1 Research design
In this study, the aim is to analyze in what way multiple source eWOM relates to opening weekend and cumulative box office and how this is different for mainstream and niche films. Because we attempt to answer our research questions through hypotheses testing, the nature of our research is deductive and we use a quantitative approach. Furthermore, all of our variables come from existing databases, and are not further manipulated by the researcher. This prevents any potential bias that can normally arise from human interactions with constructs and variables. Of course, the only exception is our film type variable, which was classified by human coders.
Regular offline WOM has always been a very difficult concept to capture effectively in a measurable variable. Earlier attempts to do so have relied on surveys and controlled laboratory experiments (Dellarocas, Awad & Zhang, 2004). Both of these methods have their own flaws. Surveys have been widely used to measure WOM in different contexts because it allows a researcher to ask respondents detailed questions about their personal communication habits and behavior. Mahajan, Muller and Kerin (1984) for instance use a survey design research to examine the effectiveness of different introduction strategies for new products in the context of positive and negative WOM. The main downside of using a survey to measure WOM is the error that lies in the self-reporting of behavior. Simultaneously, lab experiments have an issue with the generalizability of the results exactly because of the controlled, unnatural setting of the research (Dellarocas et al., 2004). The digitalization of modern society opened up possibilities to measure interpersonal communication in a more precise and authentic manner because these communications are preserved. Our database method therefore allows us to derive eWOM data on a much larger scale from multiple sources at the same time.
27
3.2 Data collection
For this research we used a dataset from the Dutch Film Distributors Association (FDN). This dataset contained all the films released in Dutch cinemas in 2014 and 2015. This initial sample of 388 films was coded by three coders as follows: 1 = mainstream, 2 = niche, 3 = unclear. The coders were instructed to classify each film based on the appeal it had on them and the way they thought the film was positioned in the market. They were instructed to classify the film as mainstream when they thought the film was meant for average, mainstream moviegoers and most related with a typical blockbuster. They were instructed to classify a film as niche when they felt like the film was targeted at a specialized target group who are interested in films that offer more quality and depth and most related to arthouse. When they did not know which category a film should belong to they were instructed to classify that film as unclear. The specific instructions for coders is listed in appendix B. Coding was done digitally in an excel file which contained the film titles, their release dates and a link to their IMDB page. When at least two out of three coders assigned a film to the same category, the film was classified as such. The films that the coders did not agree upon, meaning that it got conflicting classifications, were classified as unclear. The films that were classified as ‘3’ by a majority of the coders were given the same classification.
As stated in the section 2.6, this approach was used because earlier used methods to divide films in two meta categories (e.g. production budget, production company, type of cinema) are insufficient at this point in time. Both production companies and cinemas no longer stick to one particular type of film and are therefore no reliable sources for film type classification. Furthermore, the opinion of coders can be used as a proxy for the public opinion on film category, and in the end it is the public that fits a creative good in a particular frame based on content, appeal and social context (Currid, 2007).
28
Most of the scholars who have researched the effects of eWOM in the film industry have used a single source to collect the eWOM data from. Liu (2006), Duan et al. (2008) and Chintagunta et al. (2010) all used user-posted messages about films from Yahoo Movies as a proxy for eWOM on the films in question. At the time it made sense to use this source because it was seen as the most popular movie website, it did not require a fee for access and the structure of the website was convenient for data collection (Liu, 2006).
While Yahoo Movies removed the user-generated reviews from their website a couple of years ago, nowadays there are some other options for websites to use as a proxy for eWOM on films. Perhaps the most famous website for user-ratings of films is IMDB, which has a database of nearly every film accompanied with an average user rating. There are several reasons why such an approach is less suitable for this research. First of all, IMDB gets visitors from all over the world, while in this research I aim to analyze the impact of Dutch eWOM on Dutch box office sales. A source that would satisfy that particular condition is Moviemeter, a Dutch website on which users can post their opinions about movies and give them a one to five-star rating. However, then the question still arises to what extent the data is representative for the total amount of eWOM; an issue that every single source method has to cope with (Kim, 2014). Although Dellarocas et al. (2004) provide some survey-based evidence that online ratings can serve as a proxy for WOM, as of today it is technically feasible to capture online WOM on multiple sources simultaneously.
The eWOM data in this study is derived from Coosto, a SaaS tool that allows to search the volume and valence of written content from multiple social media sources on a given subject in a certain period of time. The tool works similar to Radian6 as used by Kim (2014), with the difference being that Coosto solely finds content generated by Dutch users. Really interesting about this tool is the wide range of popular social media sources it can use in a search. The eWOM data in this research is aggregated from the most frequently used social media channels
29
in The Netherlands: Twitter, Facebook, YouTube, Instagram and Pinterest. This multiple source approach reduces potential bias single source data potentially suffers from, considering the possible homogeneity in demographics and interests of visitors and users of specific websites (Kim, 2014). Furthermore, multiple source eWOM should give a more representative view of the overall volume and valence of content on a given subject.
For every film I used the following basic search query:
"FILM TITLE~" NEAR film* OR bios* OR zien OR gezien
The command NEAR orders the software to find content where the input term is present within a maximum of ten words of one of the terms stated to the right of the command. A star (*) next to a word is used to also find extensions of that given word. For this particular query that means that the software for instance will also find the words filmbezoek (film visit) and bioscoop (cinema). Furthermore, when a tilde (~) is placed next to a word the software will also find words that slightly deviate from the given term. This is a useful method to control for spelling errors in film names that contain words that are easily misspelled, which is very useful since most of the eWOM is user generated and by no means official company content.
The used search query probably did not find every single Dutch message on social media that exists on the subject. After all, not every post about a film has the film title within ten words of the words film*, bios*, zien (see) or gezien (seen). This addition to the query was however necessary to make sure that all of the findings were actually related to their corresponding films. When the film title Irrational Man was used as a query without the NEAR addition, Coosto found content of different subjects than the actual film while for a very specific title like Child 44 this was not the case. However, because the addition to the query was necessary to avoid an unjust volume of measured eWOM for certain films, the addition was used for all of the films in the sample in order to avoid biased results.
30
In line with my expectations beforehand it appeared that, even with the addition to the query, some of the film titles failed to isolate eWOM content solely about that particular film (N = 25). Examples of such titles were Life and Dope. These were removed from the sample, leaving a final sample of 363 films. The final sample of films is listed in appendix A.
3.3 Dependent variables
The goal of this study was to analyze the effects of fixed, moderating and control factors on the commercial performance of films. The commercial performance of films was measured by box office data. We derived all the box office data from FDN dataset. This dataset listed both the opening weekend and the total box office performance. The opening weekend performance of a film was defined as the revenue of the Thursday, Friday, Saturday and Sunday in a film’s opening week, while total box office constitutes the total cumulative revenue a film has made in Dutch cinemas. Earlier research suggests different effects of independent variables on opening weekend and total box office, especially when taking different film types into account (e.g. Gemser et al., 2007). Therefore, both variables were included in our dataset as OPENBOX and TOTBOX.
3.4 Independent variables
Earlier findings suggested that the volume of eWOM messages peaks around the theatrical release of a film (Kim, 2014; Liu, 2006). Because of this assumption, for every film in the final sample eWOM volume and valence was collected from the eight weeks before the release up to one day before the Dutch premiere date and from the eight weeks after the release, starting on the day of the premiere. We did not include eWOM content from later than eight weeks after the premiere because it would be likely that the majority of that content would not affect box office performance for two reasons. First, the amount of eWOM quickly decreases after the
31
premiere date and after six to eight weeks the volume is almost equal to zero (Liu, 2006). Second, on average films in Dutch cinemas are no longer shown than two months, so any eWOM that would come up after this period of time would not have had any effect on box office figures (BNR, 2011).
All of the eWOM data is derived from Coosto. The software states a summary of the findings with the total number of found messages, the amount of positive messages and the amount of negative messages, leaving the remaining messages as ‘neutral’. The valence of messages is automatically coded by Coosto’s algorithm. An example for pre- and post-release output for a film is presented in appendix C. By manually changing the search period for each film we were able to capture the volume and valence of both pre-release eWOM (resp. PRE_VOL; PRE_POS_PERC; PRE_NEG_PERC) and total eWOM (resp. TOT_VOL; TOT_POS_PERC; TOT_NEG_PERC). We transformed the valence variables into percentages using the function compute variable in SPSS.
3.5 Moderating variable
It was hypothesized that that the potential effects of eWOM would be different for mainstream and niche films. Although the moderating effect of this variable had not been directly tested before in this specific context, there are studies that point at the differences between types of films and the way they are affected by other external parties (e.g. Gemser et al, 2007; Reinstein & Sneyder, 2005).
After the coding process was done, the moderating variable TYFILM was included in the dataset and was given a value of 1 for mainstream, 2 for niche and 3 for unclear. In order to use the film type variable as an interaction term in a regression analysis, dummy variables had to be made for the mainstream and niche categories. In the first one all the mainstream films
32
were given the value 1 while the rest was labeled as 0 while in the latter the same was done with niche films. The unclear films were used as a reference category.
3.6 Control variables
There are several factors that are useful to control for when analyzing the impact of eWOM on film box office. The number of screens a film is released on is arguably the most obvious one of them. After all, a film that is shown on more screens is able to draw more visitors and can therefore generate a higher revenue. In numerous earlier studies the variable has proven to be an important predictor of box office performance (e.g. Gemser et al, 2007; Liu, 2006, Basuroy et. al, 2003). The number of screens were derived from the FDN dataset and included in this study as the continuous variable SCREENS.
In order to run valid quantitative analyses, it was important that the film sample was large enough. We therefore chose to take the whole population of films released in Dutch cinemas in the years 2014 and 2015 and combine those two databases in one, large dataset. This dataset therefore contained both Dutch and foreign productions. This was used as a control variable because it can be argued that films of Dutch origin generate more eWOM in the Netherlands. All of the films in the final sample were manually coded as a domestic (1) or international production (0), making up the dichotomous variable DOMESTIC.
Another factor that could have possibly affected box office are the opinions of film critics. In line with earlier research on eWOM (Liu, 2006; Duan et al., 2008; Chintagunta et al., 2010), the aggregated opinion of film critics where therefore also included in the dataset as control variable CRITIC. For international productions this data was derived from the website Metacritic.com, which aggregates a large number of reviews from US critics and states the average value judgement as a number between 0 and 100. Because only a few of the Dutch productions were featured on this website, we manually calculated the average critic rating for
33
those movies by using four well-known sources for professional film reviews in The Netherlands (De Volkskrant, Nu.nl, De Telegraaf, Het Parool). Because these sources all used a 1-5 star rating system, we manually recoded the Dutch ratings into the same scale as the international releases by using the following formula: amount of stars * 20.
3.7 Method of analysis
All the data was collected in an Excel file and exported to SPSS statistics version 20 in order to execute the statistical analyses. There were no missing values in the dependent, independent and moderating variables. Almost all of the control variables were also free from missing values, with the exception of CRITIC. For a few films it appeared that there were not enough critic reviews available with an explicit stated value judgement. Because the number of films for which this was the case was so small (N = 4), and the concerned variable was not of crucial importance to the study, in all of the conducted analyses pairwise deletion was used. This means that the data of the four films that had a missing value was still used in the analyses (Peugh & Enders, 2004).
We ran descriptive statistics on all of the variables in the dataset and checked for significant differences between the three different film types using one-way ANOVA’s. To check exactly which of the three groups’ means differed significantly from each other we ran additional post-hoc tests. Almost all of the variables had a very high standard deviation relative to their mean. This indicated a wide spread of observations within those variables. Consequently, when running the ANOVA’s for each of the displayed variables it appeared that Levene’s test for equality of variances was significant for all the variables except the control variable CRITIC. We therefore used Welch’s robust F test for every variable except CRITIC because this test is able to cope with the combined effects of unequal sample sizes and
34
heterogeneity of variances. For the same reasons we used Tukey’s post-hoc test for CRITIC and the Games-Howell post-hoc tests for the rest of the variables (Clinch & Keselman, 1982).
Because both the independent and the dependent variables were measured on a continuous scale, the hypotheses were tested using multiple regression analyses. The use of multiple regressions allowed to check for the effect of multiple independent variables on a dependent variable in different steps and also allowed for an interaction-effect check. This was done using two separate hierarchical regression analyses, one for opening weekend box office and one for cumulative box office, with three levels each. Hierarchical regression analyses were particularly useful in this study because it allowed us to isolate the effects of the eWOM variables after controlling for other variables (Woltman, Feldstain, MacKay & Rocchi, 2012). Exploration of the variables showed that both the dependent variables, all the eWOM variables as well as the control variable SCREENS had a high, positive skew and were not normally distributed. While at least for the independent variables normality is no necessity, the cone shaped residual plots of the regressions visually showed that the skewed variables caused heteroscedasticity issues. This appears when the variances of a variable highly differ among levels of another variable that predicts it. In combination with an unequal sample sizes this can lead to invalid analyses results and the drawing of false conclusions (van Peet, Namesnik & Hox, 2012). To overcome this problem, all the skewed variables were transformed using natural logs which removed both the heteroscedasticity and skewness issues. Because some films had zero negative eWOM and log transformations do not work with values of zero, all the valence variables were transformed using the formula: lnX = ln(x+1) (Berry, 1987). The new variables were included in the dataset as lnOPBOX, lnTOTBOX, lnPRE_VOL, lnPRE_POS_PERC, lnPRE_NEG_PERC, lnTOTVOL, lnTOT_POS_PERC, lnTOT_NEG_PERC and lnSCREENS.
35
The downside of using log transformations was that the interpretation of the regression results became trickier. The interpretation of the unstandardized B coefficients of the log transformed independent variables was different than the B´s of the untransformed variables because they were no longer shown in their original units (LaLonde, 2005). However, the betas of independent variables in the same regression analysis could still be used to compare the strength and direction of the effects of the different independent variables because beta coefficients measure the effect size of variables in terms of standard deviations (van Peet et al., 2012). We therefore chose to report only the standardized beta coefficients in all of the regression tables.
A set of interaction variables was computed in SPSS to check for interaction effects. This was done by multiplying all of the eWOM variables with both the MAINSTREAM and NICHE dummy variables, using the unclear films as reference category. This lead to a total of twelve interaction variables. However, it appeared that for both of the film types all of the newly created interaction variables highly correlated with each other (lowest correlation r = 0.78, p < 0.01). This caused multicollinearity issues as some of the variance inflation factors (VIF) values were way above the generally accepted limit of 10 (lnOPBOX highest VIF = 33.28; lnTOTBOX highest VIF = 33.97) (Belsley, Kuh, & Welsch, 1980). To overcome this problem, we chose to only include one interaction variable per type of film, and use this as a proxy for the other interaction variables. For both mainstream and niche films we used the interaction with lnPRE_VOL, as this variable had the highest correlations with the other variables (resp. mean r = 0.92 and mean r = 0.91).
To further explore the interactions between the different film types and the eWOM variables, we ran additional linear regressions for each of the three groups separately. We used SPSS’s ‘split file’ function to compare the results of our regressions for each of the three groups. This method was earlier used by Gemser et al. (2007) to determine the different effects of critic
36
reviews on opening weekend and cumulative box office for different types of films. Split file regressions do not allow for the regression coefficients to be compared across groups because it in fact runs separate analyses for each of the groups. However, when the sample size is large enough it can be a useful method to get a clearer overview of potential different impacts of independent variables on a dependent variable for multiple groups.
4. Results
This chapter gives an overview of the results of the statistical analyses. First, the descriptive statistics of the dataset are presented. Second, the correlations between the variables in this study are shown and discussed. The chapter is concluded with the results of both the hierarchical and the split file regression analyses.
4.1 Descriptive statistics
The 363 films in our final sample generated an aggregated grand total of €84.483.466 and €461.567.403 in opening weekend and cumulative revenue respectively. The film City of Violence had the lowest opening weekend with €8614 while Spectre’s €3.377.404 opening weekend was the highest of all. The lowest cumulative box office in our sample was the €22.712 acquired by Victoria, with Spectre again topping the list with a total box office result of €20.427.555. Interesting to note is that City of Violence and Victoria were labeled as niche and Spectre as mainstream. Furthermore, 56 of the 363 films (15.4%) were domestic productions. Niche films received an average 13.8 and 15 points higher critic rating than mainstream and unclear films respectively.
In total, the data of 368.502 eWOM messages have been included in this study. Most of the messages were posted after the premiere date, with the total pre-release messages being 139.838 (37.9%).
37
Table 1 shows a descriptive summary of the variables in the final sample. The largest part of the films in our sample were labeled as mainstream (43.8%), followed by niche (28.7%) and unclear (27.5%). The table shows that most of the released films were foreign productions, that they were released on an average of just under 70 screens and received an average critic rating of 57.6. An average film generated a cumulative revenue of approximately 5.5 times its opening weekend revenue. What struck about the eWOM variables was that the vast majority of the messages were generated during the post-release period and that on average, positive eWOM outnumbered negative eWOM by a landslide both in the pre-release period and in the total period of sixteen weeks the data was collected.
The ANOVA results show that for every variable with the exception of DOMESTIC and TOT_POS_PERC there were significant differences between means of the three types of films. In this study, we were particularly interested in the differences between mainstream and
Table 1 Descriptive statistics and ANOVA results
Total (N = 363) Mainstream (N = 159) Niche (N = 104) Unclear (N = 100) ANOVA
M SD M SD M SD M SD F DOMESTIC 0.15 0.36 0.20 0.03 0.10 0.03 0.14 0.04 2.96t SCREENS 69.6 46.2 101.1 46.5 36.6 23.3 53.8 28.2 110.20** CRITIC 57.6 17.7 54.0 16.4 67.8 17.9 52.8 16.3 27.36** OPBOX € 232.737 € 362.492 € 417.598 € 481.096 € 70.271 € 69.373 € 107.771 € 87.695 42.99** TOTBOX €1.271.536 €2.185.559 €2.278.343 €2.952.309 € 422.143 € 488.492 € 554.080 € 633.641 30.34** PRE_VOL 385.2 653.8 579.6 853.9 231.4 412.5 236.14 337.898 11.32** PRE_POS_PERC 21.8% 10.3% 20.2% 9.0% 23.1% 10.2% 23.0% 11.8% 3.70* PRE_NEG_PERC 3.8% 3.7% 4.0% 3.6% 2.8% 2.7% 4.5% 4.4% 7.84** TOT_VOL 1015.2 1559.8 1556.3 2053.7 579 870.7 608.5 734.4 15.56** TOT_POS_PERC 28.0% 9.8% 27.1% 8.7% 29.5% 9.8% 28.0% 11.1% 2.06 TOT_NEG_PERC 4.5% 3.4% 4.6% 3.4% 3.5% 2.8% 5.4% 3.7% 9.39**
38
niche films. It appeared that on average, mainstream films were released on a larger number of screens, generated more opening weekend and cumulative revenue as well as more eWOM volume, both in the pre-release period and in total. On the other hand, niche films received higher critic ratings and significantly less negative eWOM in both periods.
Table 2 shows the results of the post-hoc test for each of the variables by factor TYFILM. The means of the three film types did not differ significantly from each other for every variable with a significant F result. However, for every variable with a significant F result there were significant differences between mainstream and niche films, the two groups of which the potential differences were of particular interest in this study. The unclear films differed significantly with one of the other two groups alternately. Exceptions were
SCREENS and OPBOX, where all three groups differed significantly and PRE_POS_PERC, where Welch’s F test was significant but the post-hoc tests showed that there were only marginal differences between specific groups.
Table 2 Post-hoc tests for the differences between film types
Mainstream - Niche Mainstream - Unclear Niche - Unclear
p p p Test
DOMESTIC 0.04* 0.40 0.60 Games - Howell
SCREENS 0.00** 0.00** 0.00** Games - Howell
CRITIC 0.00** 0.85 0.00** Tukey
OPBOX 0.00** 0.00** 0.003** Games - Howell
TOTBOX 0.00** 0.00** 0.22 Games - Howell
PRE_VOL 0.00** 0.00** 0.99 Games - Howell
PRE_POS_PERC 0.05t 0.08t 0.99 Games - Howell
PRE_NEG_PERC 0.01* 0.61 0.003** Games - Howell
TOT_VOL 0.00** 0.00** 0.69 Games - Howell
TOT_POS_PERC - - - -
TOT_NEG_PERC 0.01* 0.15 0.00** Games - Howell
39
4.2 Correlations
Table 3 shows the correlations between the variables. The two dependent variables lnOPBOX and lnTOTBOX had a very high positive correlation (r = 0.91, p < 0.01). The same applied for the independent variables lnPRE_VOL and lnTOT_VOL (r = 0.94, p < 0.01), lnPRE_POS_PERC and lnTOT_POS_PERC (r = 0.68, p < 0.01) and lnPRE_NEG_PERC and lnTOT_NEG_PERC (r = 0.74, p < 0.01). This can be explained by the composition of those variables: all of the ‘TOT’ variables were operationalized as cumulative indicators and therefore also contained the data of the ‘PRE’ variables. In this case the high mutual correlations were not a problem because the concerned variables were not used in the same regression analyses.
It turned out that lnTOT_VOL had a strong positive correlation with lnSCREENS (r = 0.58, p < 0.01) and both lnOPBOX (r = 0.68, p < 0.01) and lnTOTBOX (r = 0.73, p < 0.01). In any case the number of screens a film was released on seemed to be an important variable for this study due to its strong positive correlations with both the dependent and independent variables. The correlations indicate that a higher number of initial screens was associated with higher box office, higher eWOM volume and less negative cumulative eWOM.
The valence variables had interesting correlations as well. The total positive valence variable lnTOT_POS_PERC had a very weak but significant positive correlation with lnTOTBOX (r = 0.19, p < 0.01). However, this was not the case for the pre-release positive valence variable lnPRE_POS_PERC and lnOPBOX: although their correlation was very weak and not significant, interestingly enough it was also negative (r = -0.09). Surprisingly, negative valence in the pre-release period had a very weak but significant correlation with lnOPBOX (r = 0.17, p < 0.01), but not with lnTOTBOX. This could indicate that especially in the pre-release period, it would be better to receive negative eWOM than receiving no eWOM at all.
The control variable DOMESTIC was very weakly but significantly correlated with the number of screens (r = 0.17, p < 0.01), total box office (r = 0.13, p < 0.05), pre-release eWOM
