The Influence of Expert Reviews and Online Word-of-Mouth
on Weekly Box Office Revenues
in the Movie Industry
Bachelor thesis
Faculty Economics & Business Student: Julia Groen
Student number: 10003388 Date of submission: 29 June 2015 Supervisor: Dr. F. B. Situmeang
Statement of Originality
This document is written by Julia Groen 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.
Abstract
In decision making for intangible experience products such as movies, consumers are looking for quality signals. Besides expert reviews, consumers can use the upcoming phenomenon online word-of-mouth in their decision. This thesis aims to explore how expert reviews and online word-of-mouth influence a movie;s box office success during the weeks movies are in the theatre. A combined database on movies from the websites thenumbers.com and
metacritic.com between 2000 and 2014 is used for the analysis. The analysis is carried out through a fixed effects panel data analysis in SPSS. The results of the study are in line with current literature finding that expert reviews and eWOM both have a positive effect on box office revenues. Both review types have a positive influence on box office revenues during the movie’s run in the theatre. Contradictory to the expectations, expert reviews does not have a decreasing effect on box office revenues. Combining the two review types gives a different positive effect but the effect on box office revenues appears to be insignificant.
Key words: Decision making, movie industry, signalling theory, expert reviews, online word-of-mouth, box office success, database, panel data analysis.
Table of contents Abstract 3 1. Introduction 5 2. Literature review 2.1 Signal theory 7 2.2 Expert reviews 8 2.3 Online word-of-mouth 9
2.4 Comparison between expert reviews and eWOM 11 3. Conceptual framework
3.1 Box office success 12
3.2 The effect of expert reviews 12
3.3 The effect of online word-of-mouth 13
3.4 The combined influence of expert reviews and eWOM 15 4. Research design and method
4.1 Research design 16
4.2 Data collection and sample 17
4.3 Variables 18
4.3.1 Expert reviews
4.3.2 Online word-of-mouth 4.3.3 Box office success 4.3.4 Control variables 4.4 Data analysis 22 5. Results 5.1 Descriptive statistics 24 5.2 Hausman test 25 5.3 Test of hypotheses 26 5.3.1 Hypotheses 1 and 2 5.3.2 Hypotheses 3 5.3.3 Control variables 5.4 Bootstrapping 30
6. Discussion and conclusion
6.1 Discussion of unpredicted results 31
6.2 Theoretical contribution and managerial implications 33 6.3 Limitations and directions for future research 35
6.4 Conclusion 36 A. References A.1 Articles 38 A.2 Websources 43 B. Appendix B.1 Descriptive statistics 43
B.2 Fixed effects analysis 45
B.3 Bootstrap analysis 47
1. Introduction
In the summer of 2014, a Dutch movie called ‘De Overgave’ got such negative reviews from movie critics that the movie became known as the worst Dutch movie ever (NOS.nl, 21-08-2014). The effect of the bad reviews was enormous; the biggest cinema exploitant of the country did not want to screen the movie in the cinemas and almost no revenues were earned. ‘De Overgave’ provides an example of the still existing powerful role critic’s reviews can play as quality signal. However, the same summer, the American movie ‘Transformers: age of extinction’ showed that the role of professional critics can be undone by the opinions of consumers. The negative reviews did not stop the movie from becoming a commercial success (USnews.com, 30-06-2014). This thesis will focus on this interesting relationship between the two types of reviews and revenues in the movie industry.
Traditionally, consumers would rely on the movie reviews by professional critics they could read in the newspapers, see on the television or hear on the radio. However, nowadays, an enormous upswing of the online word-of-mouth puts pressure on this traditional role of experts, as recognized in the article of Brown, Broderick and Lee (2007). The upwing of the online word-of-mouth can be accounted to technological developments as the Internet. It changes the way in which consumers buy products and services. Internet provides consumers with the opportunity to exchange their ideas and advices about products (Cheung & Lee, 2012). Many scholars acknowledge that this development has great implications for marketers in how they should deal and interact with customers. Duan, Gu and Whinston (2008) even recognize online word-of-mouth as one of the most influential sources of information for customers.
The movie industry is an interesting case while movies are intangible experience products for which the quality is hard to evaluate before you have actually consumed a movie (Liu, 2006). This experimental nature of movies causes consumers to rely heavily on reviews by others before they decide to consume the product (Chakravarty, Liu & Mazumdar, 2010). According to Situmeang, Leenders and Wijnberg (2014), expert reviews and online word-of-mouth can be seen as ‘signals’ from the signal theory; information for consumers that affect their behaviour. The question is whether the online word-of-mouth is a new substitute or a complement as signal to expert reviews.
Some articles combine the two phenomena into one variable, indicating that there is no difference between the two (Hennig-Thurau, Houston & Heitjans, 2009). However, the example of the Transformers-movie points to a possible big difference between expert
reviews and online word-of-mouth because the negative expert reviews did not stop the movie from being a commercial success.
Considering the fact that marketing literature tend to believe that expert reviews and online consumer reviews are the same, it is interesting to look at whether they could be different as signalling phenomena in the light of signal theory. There are different reasons to assume that the two types of reviews differ in content and differ in their effects on movie performance during the different weeks in a movie’s run. Moon, Bergey & Iacobucci (2010) indicated that both types are fundamentally different; expert reviews specifically function as an early quality signal when consumers have not been able to see the movie yet and the experiences
influencing the reviews are different for an expert compared to a consumer. Thereby, experts try to write an independent opinion and tend to use observable measures to evaluate technical and artistic aspects, which separates their reviews from consumer reviews (Chakravarty et al., 2010). In addition to these differences, it can be expected that the effect of the two types of reviews can be different at different points in time after the release of a movie. Expert reviews tend to have more influence on the opening weekend revenues while they can signal the quality before the movie is released (Chakravarty et al., 2010). Duan et al. point at the need for longitudinal research on this topic while a research gap exists on this point (2008). Although the study of Basuroy, Chatterjee and Ravid (2003) among other studies, did focus on the time aspect by assessing the effect of expert reviews on weekly box office revenues of movies, no clear study can be found whereby the different variables are combined with different points in time. This thesis will fill the research gap in combining those ideas into the following research question: In what way does online word-of-mouth moderate the effect of expert reviews on box office success in the weeks after the release of a movie?.
The research question will be assessed by a panel data analysis on expert reviews, online word-of-mouth and box office success in different weeks. This analysis will gain new understanding in how the two signalling phenomena work, and specifically how they
influence the movie success over time. This is particularly important while the movie industry is a risky industry where knowledge about factors driving the success of a movie is crucial (Bi & Giles, 2008). The study will therefore contribute to the literature and has practical
relevance as well. While signalling by reviews from critics and consumers should be part of the marketing strategy of managers in the movie industry (Lampel & Shamsie, 2000), more insight on the effects of these signals over time will improve the marketing strategies.
The study is structured as follows. After this first introduction, the existing literature will be discussed on signal theory, expert reviews, online word-of-mouth and the differences between the two types of reviews. Third, the conceptual framework will pose the sub
questions and hypotheses. The conceptual framework is followed by the research design and method. Chapter five will report on the analyses and the results. The last chapter, chapter six presents the discussion and conclusion in which limitations and directions for future research are considered.
2. Literature review
In this section, the existing literature on the main topics of the research question will be assessed. The signal theory is an overarching theory on how consumers use signals in buying decisions. The main signals in the movie industry exists of expert reviews and online word-of-mouth. These concepts will independently be explored and then compared.
2.1 Signal theory
When individuals make decisions, their choices are affected by information (Connelly, Certo, Ireland & Reutzel, 2011). The information can be public or private and is known for being unequally divided between individuals. The first scholar mentioning this phenomenon was Spence (1973), who explained the case of hiring a new employee. The employer does not know what the quality of the different candidates is, and therefore has to rely on signals about the quality of the candidates. These signals contain of information about the candidates, like education or what former colleagues say about the candidates. The employer will use this information to reduce the uncertainty in the decision about what candidate to hire. Lampel and Shamsie (2000) took this concept further into the product sphere. They stated that the higher the level of uncertainty about the quality of a product is, the more likely that
consumers will search for information and base their purchasing decision on this information. This theory is interesting for the movie industry while movies are experience products. With experience products like movies, consumers face great uncertainty about the quality of the product until they consumed the product (Liu, 2006). De Vany (2004) adds to this
argument with calling every movie a ‘discovery’. By this, the author means that no one can be sure about whether she/he likes a movie until she/he sees the movie. According to Lampel and Shamsie, this uncertainty is caused by the intangible nature of experience products, which makes it difficult for consumers to evaluate the movie beforehand. The time pressure caused
by the relatively short period that movies are shown in theatres further increases the need for signals (Lampel & Shamsie, 2000).
The study of Ravid (1999) focussing on potential information signals in the movie industry, found that reviews were significant signals leading to revenues. Indeed, Lampel and Shamsie indicate that expert reviews could help consumers to reduce the uncertainty about the quality of the movie. Furthermore, Situmeang et al. (2014) consider both expert reviews and (online) consumer reviews to have a signalling function while they pass on information to consumers. The information signals something about the quality of a product and causes the uncertainty of consumers about this quality to decrease.
2.2 Expert reviews
According to Holbrook (1999), professional critics offer expert judgments based on specialized criteria assimilated through extensive education or training. These expert judgments of professional critics are called the expert reviews in this study. The reviews provided by the experts signal product quality and may be of help in the decisions of
consumers (Moon, Bergey & Iacobucci, 2010). The independent evaluations of experts offer consumers the crucial information they are looking for in their decision process (Lampel & Shamsie, 2000).
In the study of Eliashberg and Shugan (1997), the possible dual role of expert critics is examined. The authors differentiate between the expert as influencer and the expert as
predictor. The expert as influencer means that the expert reviews are seen as a useful advice about a product and therefore could influence the early revenues of a product. Expert reviews would not influence the revenues of a product but could predict the success of a product, according to the expert as predictor-perspective (Eliasberg & Shugan, 1997). Boatwright, Basuroy and Kamakura (2007) also acknowledge these two possible roles, and they argue that these roles especially exist in the movie industry. The reason for this would be that every movie causes multiple expert reviews and that parts of these reviews are used for
advertisement (Boatwright et al., 2007). Thereby, the fact that movies are experience goods causes the influence of expert reviews to be high while the quality of experience goods is unknown before consumption to potential consumers (Reinstein & Snyder, 2005). Potential consumers will look for information on the quality of a movie before they decide to buy a ticket for a movie. In this sense, it can be reasoned that expert reviews influence the box office revenues of a movie (Basuroy, Chatterjee & Ravid, 2003). The study of Reinstein & Snyder found that reviews during the opening weekend of a movie can influence the box
office revenues, and hereby that expert reviews can function as both influencers and predictors (2005).
Zhu and Zhang (2009) evaluated different articles on expert reviews and concluded that there is a lot of evidence for the statement that those reviews influence the buying
decisions of consumers. An early article of Sochay (1994) named expert reviews an important marketing effort. Also Ravid (1999) showed the significance of the expert reviews in financial success of a movie and especially pointed at the quantity of the reviews (1999). The study of Brewer, Kelley and Jozefowicz (2011) looked at the relationship between various independent variables on the dependent variable box office revenue. The authors found out that critics indeed do have a positive significant effect on box office revenues in the US film industry.
Despite the extensive literature about the role of expert reviews, there is a growing discussion about the relevance of expert critics (Chakravarty, Liu & Mazumdar, 2010). This discussion mainly evolves from tendency that the society is increasingly filled with
information. Technological developments, as the wide availability of Internet change the way consumers look for information and decide about what products to consume (King, Racherla & Bush, 2014). The phenomenon of online recommendations by other consumers rapidly gains importance in buying decisions (Chen & Xie, 2008). These online recommendations can be called ‘online word-of-mouth’ and seems to have great effects on movie’s box office performance (Duan, Gu & Whinston, 2008). The literature about the phenomenon ‘online word-of-mouth’ will be further explored in the next section.
2.3 Online word-of-mouth
Recent studies indicated the importance of online conversations about products: 50% of the consumers rely on online recommendations when choosing products like movies or CDs (Awad, Dellarocas & Zhang, 2004), 67% of consumer product sales are caused by online word-of-mouth (Liu, 2006) and according to an inquiry of eMarketer, 61% of the consumers always look up online recommendations before the choose to buy a product (Cheung & Lee, 2012). The online conversations fall under the heading of ‘online word-of-mouth’ (eWOM). The eWOM is the online version of the traditional word-of-mouth, which can be described as “interpersonal communications in which none of the participants are marketing sources” (Bone, 1995). The focus of this study will not be on this offline word-of-mouth while the effect of eWOM is disproportionally greater (Chen & Xie, 2008). Cheung & Lee (2012) add to this argument with saying that eWOM is faster, ongoing, better approachable and easier to study because of the observability of the messages between consumers online. The distinctive
characteristics of eWOM against traditional word-of-mouth are the computer-mediated context, the communication between a network of people, the visibility of conversations, the shared interests of the consumers engaging in the conversations and the consumer
relationships solely existing online (King et al., 2014).
There is consensus in the existing literature about the importance of eWOM in consumer decisions (Chen & Xie, 2008; Chevalier & Mayzlin, 2006; Floyd, Frehling, Alhoqail, Cho & Freling, 2014). Liu (2006) even poses that the success of movies directly results from eWOM. The meta-analysis of Floyd et al. (2014) demonstrates the specific attributes of eWOM causing its importance: it is perceived as being more powerful because of the potential worldwide impact, as a more neutral source of information because a lot of different opinions can be present online, as more convenient because of the messages written down and as a chance to control the information flow within markets by retailers. The fourth attribute presents the opportunities for managers, who can create a monitoring system to able the company to better responsive, and incorporate the online insights in the strategy
(Dellarocas, Zhang & Awad, 2007).
The effect of eWOM varies accordingly to the status of the volume or valence of the online conversations. The volume is the number of online comments, where the valence is the preference expressed in them (Floyd et al., 2014). Liu described the different effects of this attributes as follows. Volume has an informative effect on consumer awareness and valence has a persuasive effect on attitude of consumers (2006). An examination of existing literature provides conflicting evidence regarding the two effects in the movie industry. The study of Duan, Gu and Whinston points to the importance of the awareness effect driving the box office revenues while they found a significant influence of the volume of eWOM on box office revenues (2008). Other studies also indicate the explanatory power from the volume of eWOM as more important than the effect of valence (Liu, 2006; Dellarocas, Zhang & Awad, 2007; Yang, Kim, Amblee & Jeong, 2012).
However, a study of the effect of volume and valence on 148 movies concluded that valence might be more important (Chintagunta, Gopinath & Venkataraman, 2010). More extensive studies came up with nuanced results, where the effect appears to be dependent on the circumstances. For most studies, these circumstances represented different movie
characteristics. In a study using panel data analysis with weekly box office revenues, Yang et al. (2012) only found a significant effect of valence for non-mainstream movies, and found that the effect of volume is bigger for mainstream movies. A meta-analysis of 26 empirical studies found that sales elasticity for valence is almost twice a big as the elasticity for volume
(Floyd et al., 2010). It appears to be difficult to assign the effect of eWOM to the attributes while there is no consensus about this in the existing literature.
An interesting part about the valence of eWOM is the importance of the direction. Negative reviews tend to have more power than positive messages (Chintagunta et al., 2010). In their study of the book industry, Chevalier and Mayzlin (2006) indeed found that a
negative review decreased book sales by more, than a positive review would increase the book sales. This effect is called the negativity effect (Chakravarty et al., 2010). Although the negativity effect is relatively small in experience products as movies, the effect is important to taken into account when studying eWOM (Chakravarty et al., 2010).
Recapitulatory, online word-of-mouth is a growing phenomenon with a growing important role in influencing decisions of consumers (Chen & Xie, 2008). Hennig-Thurau, Gwinner, Walsh & Gremler (2004) explain the motives of consumers engaging in online WOM and hereby establish the expectation that the phenomenon is not a short-lived trend, but that it is here to stay. According to Chevalier and Mayzlin (2006), the online word-of-mouth is both substituting and complementing other types of information consumers receive about the quality of a product. However the question remains in what way and to which extent this phenomenon is substituting and/or complementing the reviews of experts. In the next section the differences between the consumer information sources will be accessed.
2.4 Comparison between expert reviews and eWOM
The study of Hennig-Thureau, Houston and Heitjans (2009) about movie sequels provides an example of a study that combines eWOM and expert reviews into one variable to measure the quality of a movie. This could be the case while both can be seen as third-party information for consumers (Basuroy, Desai & Talukdar, 2006). However, several studies provide evidence for the proposition that the two variables are different and therefore should not be combined. A study in 1999 already indicated that experts and consumers use other criteria to create their preferences for movies (Holbrook, 1999).
Moon, Bergey & Iacobucci (2010) find fundamental differences in the experience and preferences of consumers and experts. Experts tend to focus more on the measurable
characteristics in their reviews, where consumers are influenced by their personal taste (Chen & Xie, 2008). The study of Chakravarty et al. (2010) adds to this argument with the statement that expert reviews try to avoid any influence of personal taste in their aim to provide an independent opinion. This independent opinion therefore exists mainly comments on the technical and artistic aspects (Chakravarty et al., 2010). It can be questioned whether this
independent opinion is still valued by consumers because Awad et al. (2004) found that eWOM contains of more suitable information for consumers. Also expert reviews can be characterized as static, while eWOM keeps on developing and is therefore a dynamic source of information (Liu, 2006).
The differences between expert reviews and online consumer reviews exist of the different criteria used, the type of information provided and the dynamic/static character of the reviews. These differences could explain the apparently low correlation between the two phenomena (Dellarocas, Zhang & Awad, 2007). Whether the effects of the phenomena are different will be studied in this thesis by testing the derived hypotheses with a panel data analysis.
3. Conceptual framework
The research question of this thesis is: In what way does online word-of-mouth moderate the effect of expert reviews on box office success in the week after the release of a movie?. With the use of the literature review, this research question will be approached in steps led by three sub questions and six associated hypotheses.
3.1 Box office success
The financial risks in the movie industry are high causing box office success to be an important goal (Bi & Giles, 2008). Box office revenues will operationalize the variable box office success while this is a suitable dependent variable (Chevalier & Mayzlin, 2006). Also, box office revenue is a better measure for success than number of weeks that the movie remains in the theatre (Lampel & Shamsie, 2000; De Vany, 2004). The focus will be only on box office revenues. Downloads or DVDs will be excluded. This choice is made because box office revenues indicate a high involvement of choice of consumers going to a particular movie.
3.2 The effect of expert reviews
The first sub question is ‘What is the effect of expert reviews on box office success?’. The question investigates the main relationship of the research question. The relationship is
evident in current literature. Basuroy et al. (2003) studied this relationship as the basis of their study and found a significant influence of critics’ reviews. One-third of movie consumers go to a movie because of its positive reviews (Basuroy et al., 2003). While a positive relationship
was also found in the study of Reinstein & Snyder (2005), Holbrook (1999) mention that results of this kind of studies are mixed and that no clear evidence can be found.
Moon et al. (2010) found a main effect of expert reviews on the box office revenues in the first week after the release. The revenues in the first week tend to be important while they could determine the support for further distribution and thus whether the movie could become a success (Chakravarty et al., 2010). The study of Liu also point to a correlation of critical reviews and weekly box office revenues but show that the study cannot confirm the significance of this correlation (2006). A possibility could be that expert reviews only influence the box office revenues in the beginning of runtime of a movie while Basuroy, Chatterjee and Ravid (2003) argue that the influence of experts is over when the eWOM is big enough. It will be expected that the expert reviews influence box office revenues in a positive way, but that this influence will decrease over time. The two hypotheses emerging out of these expectations are as follows.
Hypothesis 1a: Expert reviews have a positive relationship with box office revenues.
Hypothesis 1b: The influence of expert reviews on box office revenues decreases in the weeks after the release.
3.3 The effect of online word-of-mouth
Several recent studies investigated the interesting phenomenon eWOM and concluded that it is an interesting tool to influence customers (Bone, 1995; Brown et al., 2007; Chakravarty et
al., 2010). The second sub question is therefore ‘What is the effect of online word-of-mouth on box office success?’. Bone (1995) points out that eWOM is both powerful in immediate and delayed product judgements meaning that reading positive reviews from consumers online causing potential consumers to consume a product and to evaluate the product higher. Dellarocas et al. (2007) go even further with saying that word-of-mouth can be called the most important determinant for the success of a movie. The literature distinguishes between two components in online word-of-mouth each with different effects (Dellarocas et al., 2007; Liu, 2006). These are volume and valence. This thesis will focus on valence because it is interesting to compare the valence of online word-of-mouth reviews with the valence of professional reviews as the study of Chakravarty et al. (2010) showed. Also, the fact that there is no current consensus on the effect of eWOM valence among scholars makes it relevant to look at the effect of the valence. Volume will be indirectly taken into account.
The existing literature shows that eWOM becomes influential when the volume of eWOM is big enough and therefore takes time before it will influence the box office revenues (Basuroy et al., 2003; Moon et al., 2010). Apparently, eWOM gains influences as the number of online consumer reviews increases; the indirect effect of volume. Indeed significant effects where found for eWOM on the aggregate and weekly box office revenue (Liu, 2006).
However, Liu (2006) explains that the most eWOM activities do happen in the first weeks after a movie’s release creating a direct competition with the expert reviews. The expectations of the positive relationship and the increasing influence on box office revenues will be tested by the following hypotheses.
Hypothesis 2a: Online word-of-mouth has a positive relationship with box office revenues.
Hypothesis 2b: The influence of online word-of-mouth on box office revenues increases in the weeks after the release.
3.4 The combined influence of expert reviews and eWOM
The existence of different expert reviews and eWOM at different moments in time could lead to different outcomes. Moon et al. (2010) found a main effect of expert reviews on the box office revenues in the first week after the release. The revenues in the first week tend to be important while they could determine the support for further distribution and thus whether the movie could become a success (Chakravarty et al., 2010). The study of Liu also point to a correlation of critical reviews and weekly box office revenues but show that the study cannot confirm the significance of this correlation (2006).
A possibility could be that expert reviews only influence the box office revenues in the beginning of runtime of a movie while Basuroy, Chatterjee and Ravid (2003) argue that the influence of experts is over when the eWOM is big enough. This statement suits the finding of Moon et al. (2010) that the eWOM lacked an effect in the first week while the number of ratings was not big enough. Apparently, eWOM gains influences as the number of online consumer reviews increase and indeed significant effects where found for eWOM on the aggregate and weekly box office revenue (Liu, 2006).
This thesis will focus on the valence of the two phenomena while no clear evidence in the literature exists about this part of reviews. The question is whether combining the effects of expert reviews and eWOM would increase the effect on box office revenues. The third subquestion is ‘How does eWOM moderate the relationship between expert reviews and box office revenues?’. A moderation effect is chosen while it can be expected that including the variable eWOM will change the relationship between expert reviews and box office revenues while online consumers have become an important signaling sources as well (Zhu & Zhang, 2009). The hypothesis to test whether this could be the case is as follows.
Hypothesis 3a: The relationship between expert reviews and weekly box office revenues is moderated by online word-of-mouth.
Hypothesis 3b: The power of the relationship between expert reviews and weekly box office revenues increases when the variable online word-of-mouth is added.
4. Research design and method 4.1 Research design
The chosen research method to answer the research question is a quantitative empirical method. A quantitative method is appropriate when theories are tested and generated by analysing numbers (Field, 2013). This study aims to test ideas about the movie industry according to the numerical outcomes of the success of movies; the revenues. The thesis will use a database on movies derived from the websites the-numbers.com and metacritic.com. This database will function as the provider of the data for the research question. The database contains information on movies released or expected in the United States of America in between 1915 and 2020. The reason to make use of the database is that it provides a great number of observations, which enhances the relevance, the reliability of the research and also the possible generalization. More information will be provided in the data collection section. The specific method chosen is panel data analysis. This analysis was chosen because it enables to have different observations for every participant over time (Hsiao, 2003). This is important because the study wants to look into the effects at different points of time after a movie is released. Previous studies that provide valuable examples of this are the study of Bond (2002) and the study of Duan, Gu and Whinston (2008). According to Hsiao (2003),
panel data analysis has a lot of advantages over other types of research methods. Comparing panel data sets with cross-sectional or time-series data sets, he finds that panel data sets contain of substantial more data points, causing the degrees of freedom to increase and the collinearity between explanatory variables to decrease. While degrees of freedom express how much observations can vary within the population and determine how well the
population value can be estimated, high degrees of freedom are favourable (Field, 2013, p. 49). Low collinearity is favourable because there is more variation in the data whereby more reliability can be achieved (Baltagi, 2008). Other benefits of panel data analysis recalled by Baltagi (2008) are that the method controls for individual heterogeneity reducing the risk of biased individuals leading to biased results, or controlling for consistency within observations about an individual of the study. Thereby, panel data analysis allows for studying dynamics of adjustment. This is especially important for this study while it wants to see whether changes in online word-of-mouth really affect the box office revenues. Connected to this is the advantage that panel data can construct and test behavioural models better if there are complicated. The buying decisions of consumers in a hypercompetitive environment like the movie industry can be characterized as a complicated behavioural situation (Lampel & Shamsie, 2000). Therefore the choice for panel data tends to be logical.
The downsides of the research method are that it is difficult and time consuming to collect the right data and the high possibility of measurement errors (Baltagi, 2008). However, these problems do not apply to the research design of this study. The right data are already collected and accessed via the-numbers.com. The accessed database is very big, and the research will only use the data of movies of which the database contains of all the needed information. The second downside also is not applicable to this research while the participants in this study are in fact not individual humans but objective objects: movies. This causes the downside to be nonexistent while problems as nonresponse, unclear questions, memory errors or interviewer effects are specific for human participants.
4.2 Data collection and sample
As already mentioned, the data collection will consist of the creation of a dataset out of databases from two sources. The first source is the website metacritic.com which provided data on expert reviews, online consumer reviews and the names of people fulfilling the roles in the movies, as directors and actors. Metacritic.com is a website that monitors forty sources of movie reviews on a precise basis and provides extensive data including mean, standard deviation and links to the original expert reviews (King, 2007).
The advantage of the use of this database is the easy access to a great amount of data on reviews. However, the information on eWOM from the databases does not contain
information on all online reviews written by consumers causing a possible underestimation in the relationship with revenues (Zhu & Zhang, 2009). This limitation cannot be undone by searching several other sources respected the time frame of this thesis, but will be taken into account when presenting the results.
Thenumbers.com will provide the data on the weekly box office revenues and the budgets of the movies. Among others, Dellarocas, Awad and Zhang (2004) also used this website to retrieve useful data. According to them, the disadvantage is that some numbers are missing for some movies. This problem will be overcome by removing the movies with incomplete data from the dataset.
This study will focus on movies released in the US between 2000 and 2014. The focus on the US market is common among scholars (Chintagunta et al., 2010; Moon et al., 2010). The choice for the US market is made for several reasons. First, data from the US movie market arenwell documented and easily accessible. Also, the movie industry in the US is interesting while it is an important and large industry with a contribution to the economy of 10,37 billion dollars in 2014 (http://www.the-numbers.com/market/, accessed at 14-05-2015). The period between 2000 and 2014 is chosen because it contains the most information on the topic. Online reviews came into importance around the year 2000 as can be seen in the literature, as well in the rise of opportunities for consumers to express their opinion online. The websites for online movie reviews started at the end of the 1990s: Yahoo!movies was launched in 1998 and Metacritic.com started in 1999. While 98% of the box office revenues are earned in the first eight weeks (Liu, 2006), it can be assumed that movies from 2014 will not receive reviews anymore at the moment of data collection in April 2015. The initial data exists of 6422 movies, 137.464 expert reviews and 188.788 online consumer reviews. 4.3 Variables
4.3.1 Expert reviews
The definition of expert reviews in this study is as described in the article of Holbrook (1999): written judgments made by professional critics. Characteristics of the expert reviews contain of the use specialized criteria assimilated through extensive education or training (Holbrook, 1999). The expert reviews in the database exist of 137.464 reviews and come from different newspapers and magazines as Los Angeles Times, New York Post, TV guide and Variety.
The variable is operationalized by looking at the scores as coded by metacritic.com. Every movie received an average expert score between 0 and 100. These average scores will be used while coding them into specific valence categories would mean losing information. If the three category coding of the study of Chakravarty et al. (2010) would be applied, movies with scores between 0-40 would all assumed to have negative expert reviews causing the information of the distinction between a movie with 20 points and a movie with 39 points to be lost. For the main analysis, a mean centered variable is created by substracting the expert review score per movie from the average expert review score from all movies. The mean centered variable will be used to analyse the interaction between the online word-of-mouth, week number and box office revenue.
4.3.2 Online word-of-mouth and interaction
Traditionally, word-of-mouth can be described as interpersonal communications between persons in which no person is a marketing source or attached to the product which is being discussed (Bone, 1995). While the persons involved in word-of-mouth are independent of the market, the word-of-mouth communications are thought of as more reliable, credible and trustworthy (Brown, Broderick & Lee, 2007). For online word-of-mouth the same description can be kept if transferred to the online version. Therefore, the description of eWOM in this study is online interpersonal communications between persons in which no person is a marketing source.
The operationalization of this variable will be according the same structure as the variable expert reviews. By using the same procedure, the two variables can be used to create an interaction variable. The eWOM scores exist of values between 1 and 10. Two variables will be created to enable the eWOM to be used in the panel data analysis. While the online consumer reviews will influence revenues only in the subsequent weeks, the variables will be transformed as ‘lags’ to take this into account. The creation of a lag means that it assigns the eWOM scores from week 1 to week 2 etcetera. The lag variable will be used to create a mean centered variable to allow for the interaction analysis with the mean centered variable of the expert reviews. The mean centered average online word-of-mouth scores are created out of the data from the database. The database exists of approximately 29 reviews per movie on average with 188.788 data points in total of online consumer reviews.
Finally, multiplying the mean centred online word-of-mouth variable with the mean centred expert review variable created the interaction variable between online word-of-mouth and expert reviews. This variable allows the analysis of hypothesis 3a and 3b.
4.3.3. Box office success
The dependent variable box office success stands for the financial success of a movie (Delen, Sharda & Kumar, 2005). According to Delen et al., there are nine categories to classify the success of a movie: the first category is named ‘flop’, for movies earning less than 1 million, and the last category is named ‘ blockbuster’, for movies earning 200 million or more (2009). Classification of success according to the revenues is common in the literature while the revenue is an ideal dependent variable (Chevalier & Mayzlin, 2006). This study only focus on box office revenues and therefore excludes revenues from other sources as downloads or DVD sales.
In this study, the box office revenues per movie per week will measure the box office success. This allows the identification of the box office revenues pattern for every movie. The chosen operationalization is common in the literature and it is known that box office revenues are reliable representations of the success of a movie (De Vany & Lee, 2000; Basuroy et al., 2003). The data for box office revenues will be arranged in a way that the information can be used as panel data per week. A script in Microsoft Access is used to create a variable in which the revenues for a movie collected at specific dates were accounted to the right week number, counted from the release. This way, every movie has data on box office revenues per week for a certain amount of weeks; differing between movies according to their success. This resulted in the variable of weekly box office revenue. Furthermore, a new variable was created as the logarithm of the weekly box office revenue variable to make the information usable for the panel data analysis.
4.3.4 Control variables
This thesis will include several variables as control variables. Control variables are variables that might influence the relationship between the dependent and independent variable. The first variable that should be included is the genre of the movie (Boatwright et al., 2007; Liu, 2006). For this variable, a dummy variable will be created in Microsoft Excel with ‘drama’ as basis and 22 genres as dummies. Every movie gets a 0 or 1 indicating that the genre is not or is applicable to the movie. More genres could be applicable to one movie. Second, the variable star power should be controlled for while numerous studies point at the possible influence on box office revenue and even word-of-mouth (Sochay, 1994; Liu, 2006, Boatwright et al., 2007; Brewer et al, 2011). With the data on the directors and
taking into account whether they were nominated, or even won an award for previous movies. The nominations and won awards signal experience and quality. Movie awards that will be included are the Oscars and the Golden Globes. The list of nominees and winners from 1980 until 2013 will be matched to each movie in the sample to see how many nominations and awards a movie got for best movie, best director, best (supporting) actor and best (supporting) actress. The third control variable is movie budget, while it is said that budgets moderate the role of expert reviews (Basuroy et al., 2003). In de database, there is information about the production budget of every movie. The movie budget will be taken into account as a ratio variable. The last control variable is age restriction. The MPAA ratings, representing the age admittance/restriction per movie, are often taken into account while they tend to influence the success of a movie as well (Sochay, 1994; Delen et al., 2005; Brewer et al., 2011). The three main categories of the MPAA ratings will be taken into account: no one 17 and under
admitted [adult category], parents strongly cautioned [teen category] and general audiences [all ages category] (mpaa.org,, 21-06-2015).
Table 1 Variable overview
Variable Name Type of data
Metascore MetascoreWeek Metascore_mc Val_expert AvgScr lagAvgScr Expert reviews Online word-of-mouth lagAvgScrWeek lagAvgScr_mc Val_eWOM
Numeric scores of expert reviews.
Numeric scores of expert reviews per week. Mean centered variable expert reviews. Valence coded categorical variable experts.
Numeric scores of online word-of-mouth. Lag variable of eWOM.
Lag variable of eWOM per week.
Mean centered lag variable eWOM per week. Valence coded categorical variable eWOM.
IntAvg_Meta
Rev_week Interaction
Box office success
Rev_week_nl
Interaction variable between lagAvgScr_mc and Metascore_mc.
Box office revenues per week.
Lag variable of box office revenues per week.
CV_Crime etc.
CV_Nom Star power
CV_Win
Number of nominations actor/actress/director. Number of won awards actor/actress/director.
Movie budget CV_Prodbudget Number representing the budget of the movie.
CV_Adult CV_Teen Age restriction
CV_Allages
Dummy variables for the three possible categories.
4.4 Data analysis
The panel data analysis will be carried out through different steps. After the data is transformed into panel data with Microsoft Access, the data will be analyzed with the statistical program R according to the basic formula of panel data analysis. This formula is yit= α*i + βixit + uit (Hsiao, 2003). In this formula y is the dependent variable, x is the
independent variable, α is a coefficient, β is a coefficient and u is the error term. The i stands for individuals in the study, and the t is the indication of the time (Hsiao, 2003). In this study, the box office success is the dependent variable and the expert reviews are the independent variable at first. The point of this study is to find out how the relationship between de dependent variable and independent variable is different for the different movies (the
individuals in this study) at different points in time (the weeks after the release of the movie). Also, it will be tested whether this relationship is different when the online word-of-mouth as coefficient β is put into the equation. The control variables that were mentioned will put into the equation as the α coefficient.
The transformed data will be put into SPSS accordingly to the formula and the
regressions will be performed. A hierarchical regression will be performed with two different levels. The first level contains of the movies and the second level contains the time, or the different weeks in this study. The first step is to summarize the data in a clear table with mean, standard deviation, minimum and maximum. While the data on movie revenues change over the different time observations, this table will show standard deviations overall, within (the same movie) and between (the different movies) to account for the possible differences. Next step is the performance of the Hausman test to check what type of panel data analysis suits the data of this study. The null hypothesis of the Hausman test is that the random effects model is the preferred model. Therefore, if the p-value is not significant, the random effects
model needs to be used. On the opposite, when the p-value is significant, the fixed effect model is the most appropriate model. After deciding on the appropriate model, the real data analysis can start. The panel data analysis will be carried out through SPSS. The data as prepared in excel should be transformed into the excel format with movie number and week number as the first two variables followed by the real variables. In this study, movie number can be seen as participant number and week number represents the time dimension.
The data will be explored by using the descriptive statistics possibilities in SPSS. The panel data analysis is carried out through linear mixed models in the analysis section. In this analysis, there are three possible frames: dependent variable, factor(s) for the class variables, and covariate(s) for the continuous variables. The dependent variable in this study is the revenue per movie per week. The factors are age restriction and genre. The main variables and the other control variables production budget and star power can be assigned to the covariates. When a fixed effects model is the preferred model, a model should be build with all the variables excluding the intercept with type III sum of squares. The estimation method is the restricted maximum likelihood (REML). This method is useful because it can
differentiate between the information within a model. This method is a well-established method for estimating parameter in a mixed linear model (Kenward & Roger, 1997). Kenward and Roger (1997) name the REML method as a more precise measure for the coefficients while it can take into account different combinations of covariance structure, design and sample size. The model statistics used for the conclusion are the parameter estimates with a confidence interval of 95%.
The conclusions of the study will be carefully reported and derived out of the results, which will be presented in clear tables. The precise report is important to increase the reliability of the study. Reliability is whether an instrument can be interpreted consistently across different situations (Field, 2013). It is important that similar results would come out, if the study would be repeated in the same way. Therefore it is important the steps taken in the study are very clear. This study can be said to be relatively reliable while the research method is standardized using a computer program and the steps taken can be exactly copied after reading this study. The reliability of the different measures will be assessed with a bootstrap analysis. The validity of this study is quite high because the instruments used to measure the concepts already proved themselves to measure what they intend to measure in existing literature as described in the beginning of the methodology section (Field, 2013). The
possibility of generalizing the results of this study will be greater than a cross-sectional study because of the great amount of data points and the high ‘N’ or number of participants
(Baltagi, 2008). However, the data only focus on movies released and reviewed in the USA. This fact decrease the generalisability because of the potential differences in the preferences and culture of consumers, as well differences in the movie industries and systems of other countries. These comments on the quality of the study will be further assessed and will be taken into account in the results and discussion section.
5. Results
5.1 Descriptive statistics
After the preparation of the data, the first step in the analysis is the exploration of the data by looking at the descriptive statistics. Table 2 provides the descriptive statistics of every variable this thesis will look at. The table provides an overview and shows how the range of the variables.
Table 2 Descriptive statistics overview
Variable N Mean value Std. Error Min. Max.
Movie_ID 60369 14 6420 Week_ID 60369 16.25 15.12 1 86 Rev_week 10007 6531494.43 18849900.88 -17888209 755849765 Rev_week_nl 9991 13.61 2.53 5.17 20.44 lagAvgScr 56192 11.72 9.47 0.25 68.75 lagAvgScr_mc 56192 6.39 6.98 0.03 57.03 lagAvgScrWeek 56192 117.42 333.94 0.39 3977.74 MetaScore 60369 58.00 19.63 1 100 MetaScore_mc 60369 16.18 11.11 0 57 MetaScoreWeek 60369 199.67 263.31 0 3519.38 IntAvg_Meta 56192 111.93 184.95 0 2794.55 Val_eWOM 26045 2.38 0.76 0 3 Val_expert 60369 2.09 0.70 1 3 CV_Prodbudget 60369 39673071.09 50088908.59 0 425000000 CV_Nom 60369 3.42 5.11 0 40 CV_Win 60369 0.88 1.69 0 15 CV_Adult 60369 0.45 0.50 0 1 CV_Teen 60369 0.35 0.48 0 1 CV_Allages 60369 0.20 0.40 0 1 CV_Drama 60369 0.48 0.499 0 1 CV_Crime 60369 0.17 0.377 0 1 CV_Comedy 60369 0.29 0.454 0 1 CV_Romance 60369 0.19 0.392 0 1 CV_Adventure 60369 0.19 0.392 0 1 CV_Fantasy 60369 0.18 0.381 0 1 CV_Mystery 60369 0.12 0.321 0 1 CV_Thriller 60369 0.13 0.344 0 1 CV_Biography 60369 0.35 0.476 0 1 CV_Horror 60369 0.04 0.205 0 1 CV_Family 60369 0.10 0.300 0 1
CV_Musical 60369 0.07 0.262 0 1 CV_History 60369 0.13 0.332 0 1 CV_Documentary 60369 0.03 0.178 0 1 CV_War 60369 0.02 0.144 0 1 CV_Western 60369 0.04 0.194 0 1 CV_Sport 60369 0.01 0.094 0 1 CV_Animation 60369 0.03 0.159 0 1 CV_Short 60369 0.02 0.147 0 1 CV_Adult 60369 0.00 0.010 0 1 CV_FilmNoir 60369 0.00 0.013 0 1 CV_News 60369 0.00 0.009 0 1 CV_Music 60369 0.00 0.006 0 1 5.2 Hausman test
The basic formula of panel data analysis has to be classified further in order to analyze the data. The extension of the formula depends on whether the coefficients of the model are random or fixed (Hsiao, 2003). Hsiao (2003) marks that the formulation of the best possible model for panel data analysis can be seen as a difficult process. A helpful tool in the choice between a random effects or fixed effects model is the Hausman test (Baltagi, Bresson & Pirotte, 2003). The Hausman test is widely used in the economic research field and contrasts the fixed effects estimator with the random effects estimator. The null hypothesis of the test states that the conditional mean of the disturbances given the regressors is zero, with the indication that the random effects model would be more consistent and efficient. The alternative hypothesis would indicate that using the random effects model would cause bias and that therefore the fixed effects model would be better suitable.
For this study, the Hausman test was performed in the statistical program R. First both the fixed effects model and the random effects model were run with weekly box office
revenue as dependent variable, week numbers and the interaction variable of eWOM and expert reviews as independent variable, and movie numbers together with week numbers as panel setting variables. The detailed report is included in appendix B5. The model provided a chisquare score of 151.86 with a value of 0.00000000000002381. While the presented p-value is lower than 0.001, the null hypothesis has to be discarded. This indicates that a random effects model would be biased and that the fixed effects model is therefore the preferred model.
The intuition behind the assumption of the fixed effects model is that there are differences within the individuals (the individual movies in this study) that may affect the outcome variables. Unlike the random effect model, the fixed effect model controls for this possible bias by removing the effect of the individual characteristics (Torres-Reyna for
Princeton, 2007). The formula that will be used for the panel data analysis in this study will be as follows.
Yit = β1Xit + αi + uit
Y represents the dependent variable with the dimensions entity (movie ID in this study) and time (weeks in the study). X represents one independent variable with β as coefficient. The α term is the unknown coefficient for every entity and the u is the error term.
5.3 Test of hypotheses
5.3.1 Hypotheses 1 and 2 – the effect of eWOM and expert reviews
In this section the analyses in SPSS will be performed in order to test the hypotheses of this thesis. For the analysis, the main variables and control variables were combined into one model but the control variables will be discussed in the next section. Table 3a reports the results on the analysis of the main variables.
Table 3a Estimates of Fixed Effects model - main variables
Variable Coefficient Std. Error T-statistic
Week_ID (.188796)*** 0.0160 -11.790 lagAvgScr 0.161856*** 0.0112 14.447 Metascore 0.010983*** 0.0013 8.166 lagAvgScr_mc*eWOM_week 0.008458*** 0.0017 5.087 Metascore_mc*eWOM_week 0.000883 0.0005 1.658 IntAvg_Meta 0.001946* 0.0009 2.100 IntAvg_Meta*eWOM_week (.000107) 0.0000 -1.407 Note. DV=Rev_week_nl, *p<0.05, **p<0.01, ***p<0.001
The first hypothesis (1a) predicted that expert reviews would have a positive relationship with box office revenues. The results of the model indeed show a positive effect with a coefficient of 0.010983, significant at the 0.001-level. This indicates that if the scores of the expert reviews increase with 1, the box office revenues would increase with 1,0983%. Although the effect seems quite small, the found effect confirms the hypothesis. To access why the effect is rather small, the next hypotheses including the time dimension is important to study the effect more closely.
The analysis shows an insignificant positive effect of expert reviews on box office revenues across weeks (β= 0.000883, p>0.05). In contradiction to hypothesis 1b, the result
implies that the effect of expert reviews would become bigger during the run of a movie in the theatre. However, a p-value of 0.097 is reported causing the effect only to be significant at a 10%-level. Significance at a 10%-level is less strict than significance at 5, 1 or 0.1%-levels. The conclusion that hypothesis 1b is disconfirmed by the results should therefore be drawn with caution.
The first found effect of online consumer reviews (lagAvgScr) is both positive and significant, confirming hypothesis 2a. Hypothesis 2a evaluates the proposed effect between online word-of-mouth and box office revenues. The coefficient of this relationship seems to be 0.1619 (p<0.001) confirming the positive relationship between eWOM and box office revenues in this study. The addition of the time dimension into the analysis reports a positive effect as well. The effect is reported as highly significant (β= 0.008458, p<0.001).
Apparently, the effect on box office revenue increases as the weeks pass by and more online word-of-mouth is available. This is in line with the literature-based expectation and confirms hypothesis 2b. The result points in the direction of an increasing effect of online word-of-mouth, possibly taking over the effect of expert reviews after the early weeks of the movie’s run. An interesting fact seems to be that the total positive effect of online-word-of-mouth on box office revenues is bigger than the positive effect of eWOM as the weeks pass by. 5.3.2 Hypothesis 3 – the combined effect
As can be seen in table 3a, the coefficient of the interaction effect of eWOM and expert reviews on the box office revenues is 0.001946 (p<0.05). Our hypothesis 3a predicted that the relationship between expert reviews and weekly box office revenues would be moderated by online word-of-mouth. A comparison between the coefficient of the interaction effects test (β= 0.001946) and the coefficient of the test on expert reviews and weekly box office
revenues (β= 0.010983) shows that the coefficient is indeed changed because of the addition of eWOM into the model. This change indicates that there indeed is an influence of the eWOM on the main relationship. The influence can be said to be moderating while the addition of the variable affects the relationship directly, instead of the relationship being affected indirectly through eWOM, which would have indicated a mediating effect.
For the assessment of hypothesis 3b, first a graph is created to get an idea of the data. Graph 1 show that the combination of low (meaning low score for valence of) expert reviews and low eWOM has the smallest effect on weekly box office revenues. The effect becomes bigger as the low eWOM is combined with high expert reviews. However, the biggest effect can be found for the combination of high eWOM and low expert reviews.
Low expert reviews in combination with low eWOM show the lowest effect on weekly box office revenue. Low expert reviews in combination with high eWOM show the highest effect on weekly box office revenues. If high eWOM is combined with high expert reviews, the effect goes down but is still bigger than any effect in case of low eWOM. The representations in the graph seem to fit the hypotheses. The statistical analysis will have to show whether the proposed relationships in the graph can be concluded from the data.
Graph 1 Relationship between expert reviews and online word-of-mouth on box office revenues
The analysis on hypothesis 3b is meant to provide clarity on the most complicated relation in this thesis. It states that the power of the relationship between expert reviews and weekly box office revenues would increase when the eWOM is added. Contradictory to the expectation, the coefficient of the model seems to decrease when the variable eWOM is added from 0.01093 (p<0.001) to 0.001946 (P<0.05). The result implies that the effect of expert reviews and eWOM together diminish the effect on the box office sales. Nonetheless, the statistical test of the interaction between expert reviews and online word-of-mouth delivers an insignificant negative coefficient. The analysis cannot provide convincing evidence on the moderating effect of online word-of-mouth on the relationship between expert reviews and box office revenues.
5,72 5,74 5,76 5,78 5,8 5,82 5,84 5,86 5,88 5,9 5,92
Low Expert reviews High Expert reviews
D ep en d en t var iab le
Low Online word-of-mouth
High Online word-of-mouth
5.3.3 Control variables
Table 3b presents the results of the fixed effect analysis of the control variables. The first control variable is production budget. The analysis shows a significant effect with a
coefficient less than 0.0000 (p<0.001), pointing at evidence for the effect of the production budget to be zero.
The second variable to control for is star power, consisting of two sub variables concerning movie awards. The nominations of a movie including previous nominations of actors and directors, appears to have a moderate effect on weekly box office revenues (β= 0.0309, p<0.001). However, no significant effect for winning movie award(s) is found (β= -0.0114, p>0.05). It could be that the nomination already creates the star power among
directors and actors. Winning an award creates no extra star power according to these results. The conclusion appears to be that award nominations should be taken into account when analysing weekly box office revenues.
Thirdly, age restriction is analysed to check for effects. The three variables representing the three age categories according to the MPAA show different effects. If a movie belongs to the category of 21up, this would increase the box office revenues significantly at a 0.1%-level (β= 5.5015). A categorization for the age category teen has a significant negative effect on the revenues with a coefficient of -1.0620 at a 0.1%-level. The last age category shows a non-significant effect (β= 0.0872, p>0.05). Although not all categories show a significant effect, it is clear that age restriction could be a variable influencing box office revenues.
The last control variable included is movie genre. This variable is divided over 23 sub variables, each representing a specific genre matched with every movie to check whether the genre applied to a movie. In summary, the analysis gives significant coefficients for fourteen genres, non-significant coefficient fro five movies and four variables are said to be redundant. The exact scores per genre can be found in table 3b. Film Noir is the genre with the biggest reported positive effect (β= 4.5119, p<0.01) and Musical with the biggest negative effect (β= -1.3917, p<0.001). The big differences found in the coefficients of the different genres reveal an interesting effect within this control variable. Therefore, the conclusion on the variable genre would be that the effect of genres on box office revenues has to be acknowledged though specific sub variables.
Table 3b Estimates of Fixed Effects model - control variables
Variable Coefficient Std. Error T-statistic
CV_Prodbudget 1,423E-8*** 0.0000 28.276 CV_Nom 0.0309*** 0.0064 4.834 CV_Win (0.0114) 0.0181 (0.063) CV_21up 5.5015*** 1.4363 3.830 CV_Teen (1.0620)*** 0.1507 -7.048 CV_Allages 0.0872 0.1564 0.558 CV_Drama 0.6859*** 0.0501 13.696 CV_Crime (0.1324)* 0.0660 -2.007 CV_Comedy (0.0165) 0.0514 (0.322) CV_Romance (0.3510)*** 0.0594 -5.913 CV_Adventure 0a 0a 0a CV_Fantasy (0.3593)*** 0.0681 -5.274 CV_Mystery 0.3866*** 0.0754 5.126 CV_Thriller (0.0815) 0.0753 -1.083 CV_Biography (0.1438)* 0.0617 -2.332 CV_Horror (0.3705)*** 0.1044 -3.549 CV_Family 0.2794** 0.0866 3.227 CV_Musical (1.3817)*** 0.0977 -14.140 CV_Scifi (0.3018) 0.1564 -1.930 CV_History 0.6076*** 0.0752 8.083 CV_Documentary (0.6969)*** 0.1284 -5.426 CV_War 2.4463*** 0.1585 15.432 CV_Western 0.9439*** 0.1480 6.376 CV_Sport 0.1578 0.2711 0.582 CV_Animation (0.4740)** 0.1415 -3.350 CV_Adult 0a 0a 0a CV_FilmNoir 4.5119** 1.3207 3.416 CV_News 0a 0a 0a CV_Music 0a 0a 0a
Note. DV=Rev_week_nl, *=p<0.05, **=p<0.01, ***=p<0.001, a=parameter is
redundant
5.4 Bootstrapping
A bootstrap analysis is performed to check whether our data is stable. The analysis is performed in the mixed model analysis in SPSS. The bootstrap method used is the random number generator in a Mersenne Twister-setting. The bootstrapping took 840 random samples from the data to see if the results of those samples are comparable to the results of the main model. The results of the bootstrap are presented in table 4.
Table 4 Bootstap for Estimated of Fixed Effects
Variable Coefficient
Coefficient
MM Bias Std. Error Std. Error MM
Week_ID (.188796)** (.188796)*** 0.00040 0.0129 0.0160 lagAvgScr 0.161860** 0.161856*** 0.00190 0.0135 0.0112 Metascore 0.010983** 0.010983*** (0.00008) 0.0013 0.0013 lagAvgScr_mc*eWOM_week 0.008458** 0.008458*** 3,98E-05 0.0013 0.0017 Metascore_mc*eWOM_week 0.000883 0.000883 (0.00004) 0.0005 0.0005 IntAvg_Meta 0.001946 0.001946* (0.00016) 0.0012 0.0009 IntAvg_Meta*eWOM_week (.000107) (.000107) 1,97E-05 0.0001 0.0000
Note. Based on 849 samples. DV=Rev_week_nl, *p<0.05, **p<0.01, ***p<0.001
Next to the results of the bootstrapping, the results of the main model are presented to enable a clear comparison. As can be seen in the table, the results of all the coefficients are identical. Minor differences can be found in the standard error but this does not change the significance of the coefficients. With the replication of the analysis bootstraps create samples to see whether the estimations of the parameters are precise (Lange, Little & Taylor, 1989). In their analysis Lange et al. show also that it is normal that standard errors of bootstrap samples are somewhat different from the standard errors in the main analysis.
6. Discussion and conclusion
6.1 Discussion of unpredicted results
The first unpredicted result in this study is that no significant influence of expert reviews is found as the weeks go by. The expectation in this study was that the influence of expert reviews would start high but would decrease over the weeks when the online word-of-mouth would start to increase. The findings in this study provide prudent evidence for the opposite while an positive effect was reported at a 10% significance level.
One reason for this finding could be the Information Overload concept. This concept describes the tendency that consumers make less accurate decisions when they face too much information (Eppler & Mengis, 2004). Important determinants of when information becomes too much are time and information characteristics. Applied to the situation of consumers evaluating a movie the Information Overload concept means that the processing capacity available in the decision period and the attributes of the reviews such as its intensity, will determine whether a consumer feels overwhelmed by the information. While movies are intangible products, consumers will go through of process of looking for informative signals before deciding to consume the product (Liu, 2006). It can be argued that a consumer will feel easily overwhelmed by the amount of information of online consumer reviews while the
amount on Internet will increase as the weeks go by. This effect could be enhanced by the fact that consuming a movie is a relatively an easy decision. It could be that consumers soon reach a state of information overload and therefore, they abandon the online consumer reviews to focus on the expert reviews. This would be in line with the findings on the effect of expert reviews, which appears to increase as the weeks go by. The article of Jones, Ravid and Rafaeli (2004) about the information dynamics in online interaction spaces confirms that consumers end their online participation when the mass interaction grows. A more elaborate analysis has to be undertaken to see whether the Information Overload concept is indeed present for consumers regarding the movie industry. This direction and other possible directions of future research will be mentioned in section 6.4.
The second plausible reason for the unexpected finding of increasing influence of expert reviews on weekly box office revenues is the twofold impact of expert reviews as recognized in the literature. Reinstein and Snyder point to the difference between the
influence effect and the prediction effect (2005). The influential role signifies the function of opinion leader; experts are seen as appropriate source of information and advice, which leads to their influence on the box office revenues. The other role as predictor is the role of leading indictor; experts are a representation of the audiences and therefore, they can envision the box office success (Eliashberg & Shugan, 1997). While this thesis does not separate between these roles, it could be that the influence-role is apparent in the first week(s) of the movie’s run and that the predictors-role then takes over. In the study of Reinstein and Snyder (2005) this appeared to be the case; they claim that expert reviews only have a prediction effect after the first weekend of a movie’s run. It could be the case that both effects are apparent in this study, causing positive effect of expert reviews on box office revenues due to the influence-effect. Additionally, it causes a positive effect of expert reviews on box office revenues due to the predictor-effect. In section 6.3, a new study design will be proposed to control for both roles in future research.
A third reason for the increasing effect could be the use of the expert reviews in advertisements in the following weeks. It is known that the largest part of the advertising budget for movies is spent before the release of a movie (Dellarocas, Zhang & Awad, 2007). However, if movies are received well, the advertising campaigns are usually extended (Dellarocas et al., 2007). In those extended advertising campaigns, quotes from expert reviews are often used as an objective quality signal to potential consumers (Boatwright, Basuroy & Kamakura, 2007). In this way, it can be argued that the effect of expert reviews is extended via the advertising campaigns of successful movies.
