exploring the “twitter effect:” an investigation of the ... · —the cluetrain manifesto...
TRANSCRIPT
1
Exploring the “Twitter Effect:” An Investigation of the Impact of Microblogging Word of
Mouth on Consumers’ Early Adoption of New Products
Thorsten Hennig-Thurau
Marketing Center Muenster University of Muenster
48143 Muenster, Germany & Cass Business School, City University London
London EC1Y 8TZ, UK Phone (+49) 251 83 29954 Fax (+49) 251 83 22032
Email: [email protected]
Caroline Wiertz Cass Business School
City University London London EC1Y 8TZ, UK
Phone (+44) 20 7040 5183 Fax: (+44) 20 7040 8262
Email: [email protected]
Fabian Feldhaus Marketing Center Muenster
University of Muenster 48143 Muenster, Germany Phone (+49) 251 83 29954
Email: [email protected]
Acknowledgments: The first and second author contributed equally to the project. The authors thank Andre Marchand as well as the participants of research seminars at Cass Business School, the University of Muenster, the University of Hamburg, the Technical University of Munich, HEC Paris, and the 2010 UCLA/Bruce Mallen Scholars and Practitioners Workshop in Motion Picture Industry Studies for their constructive criticism on previous versions of this manuscript. They also thank Benno Stein and Peter Prettenhofer for their help with the WEKA analysis, Mo Musse and Peter Richards for their IT help and Chad Etzel from Twitter for supporting the data collection. Finally, the authors are grateful for research funds provided by Cass Business School and City University London that supported this project. Keywords: Word of mouth communication, microblogging, Twitter, early adoption.
Working Paper, March 5, 2012
2
Exploring the “Twitter Effect:” An Investigation of the Impact of Microblogging Word of Mouth on Consumers’ Early Adoption of New Products
ABSTRACT
Microblogging word of mouth (MWOM) through Twitter and similar services constitutes a
new type of word-of-mouth communication that combines the real-time and personal influence
of traditional (offline) word of mouth (TWOM) with electronic word of mouth’s (EWOM)
ability to reach large audiences. MWOM has the potential to increase the speed of dissemination
of post-purchase quality evaluations from consumers and thus has been argued to affect early
product adoption behaviors. For industries that exploit information asymmetries between
producers and consumers when releasing new products, such a “Twitter effect” would threaten
existing business models. This study develops a conceptual model of the impact of MWOM on
early product adoption, including possible moderating forces, and tests it in the context of the
motion picture industry. Studying 105 movies that were widely released in North American
theaters between October 2009 and October 2010, and all 4 million MWOM messages about
them sent via Twitter on their respective opening weekend, the authors find evidence of the
“Twitter effect” and identify boundary conditions. With a matched sample of 105 movies
released in the pre-MWOM era, the authors also demonstrate that the spread of quality-related
information by consumers through MWOM is indeed the cause of this effect. The authors
discuss notable implications for managers of experiential media products and word-of-mouth
scholars.
3
Exploring the “Twitter Effect:” An Investigation of the Impact of Microblogging Word of
Mouth on Consumers’ Early Adoption of New Products
“The Internet is enabling conversations among human beings that were simply not possible in
the era of mass media.”
—The Cluetrain Manifesto (Levine et al. 2000, p. XXII)
Recent world events, such as Iran’s last presidential election and the Arab Spring movement,
have compellingly demonstrated the power of microblogging for the rapid spread of information
among networked individuals (Kaplan and Haenlein 2011). Microblogging refers to the
broadcasting of brief messages to some or all members of the sender’s social network through a
specific web-based service. Although various microblogging services exist, Twitter has become
synonymous with the concept; it alone boasts more than 100 million active users (October 2011)
and processes approximately 250 million messages every day, more than 40% of which are
posted “on the go” using mobile devices (Parr 2011a, 2011b).
For marketing, microblogging enables a new type of word-of-mouth communication for
which we introduce the term microblogging word of mouth (MWOM). Such MWOM combines
elements of both traditional (offline) word of mouth (hereafter: TWOM; Katz and Lazarsfeld
1955) and electronic word of mouth (EWOM; Hennig-Thurau et al. 2004; Liu 2006): It reaches a
potentially very large number of consumers with a single message (which EWOM can, but
TWOM cannot), enables real-time information sharing among consumers from any place, and
relies on a personal connection between the sender and receiver (both of which TWOM can, but
EWOM cannot). This unique combination of characteristics implies that MWOM can reach a
vast number of consumers at unprecedented speed.
4
The aim of this research is to investigate whether MWOM, due to its unique characteristics,
influences the success of new products by shifting consumers’ early adoption behaviors. Such a
“Twitter effect”1 (Corliss 2009) would have strong implications for products that depend on
instant success upon their release – at a point in time when consumers are unable to judge their
“true” quality and must make adoption decisions mainly on the basis of promotional material.
Examples of such products include experiential media products (e.g., movies, music, electronic
games), but also products that benefit from a hyped release (e.g., Apple’s iPhones and iPads).
Among motion picture industry experts and journalists, for example, proponents of the
“Twitter effect” blame MWOM for the immediate failure of multimillion projects such as Brüno
and G.I. Joe, as well as for the unexpected opening successes of Transformers and The Karate
Kid, despite their negative reviews by professional critics (e.g., Corliss 2009; Lang 2010). If
MWOM does affect the early adoption of new products, investments in risk-intense products
would become even riskier and less attractive, because MWOM threatens to decrease the share
of revenues that remain unaffected by consumers’ quality perceptions of the product. Although
this goes beyond the scope of this research, the importance of action-based cascades, such as
opening weekend box office lists for movies, implies that the “Twitter effect” could also
influence subsequent revenues, because a large number of consumers base their purchase
decisions on such quality-neutral information (Bikhchandani, Hirshleifer, and Welch 1992).
However, despite anecdotal evidence in support of the “Twitter effect”, it is far from certain
that MWOM exerts such an impact. Other industry insiders and journalists (e.g., Goldstein and
Rainey 2009) question its existence, and a large-scale study of the media habits of moviegoers
reveals that approximately half of the respondents self-report that they rely only on TWOM
1 This effect, though named in reference to Twitter as the dominant market leader, is not solely exerted through Twitter’s service but refers to microblogging’s impact in general.
5
when making purchase decisions, but ignore MWOM (Lang 2010). Moreover, MWOM’s very
short message content has been criticized for limiting the amount of information that can be
transmitted and thus limiting the impact of that information (Dugan 2010; Goldstein and Rainy
2009; Lang 2010).
In this study, we address this debate and advance the word-of-mouth literature in marketing.
Specifically, we aim to make three contributions: First, we introduce and conceptualize the
concept of MWOM and position it in relation to TWOM and EWOM. Second, we develop a
conceptual model of MWOM’s impact on the early adoption of new experiential media products,
including its boundary conditions. Third, we test our model empirically for the early adoption of
new movies, drawing on a unique data set of all MWOM messages sent via Twitter that
pertained to 105 widely released movies during their respective North American opening
weekend. Based on sentiment analysis and seemingly unrelated regression analyses, we find
support for the “Twitter effect.” Comparing the findings with a second sample of matched
movies from the pre-MWOM era confirms that the “Twitter effect” is indeed a result of the early
availability of post-purchase quality-related information from consumers, as enabled by
MWOM.
THEORETICAL BACKGROUND
One of marketing’s law-like generalizations states that word-of-mouth communication is a
key information source for consumer decision making (Arndt 1967; Godes and Mayzlin 2004).
Over the years, different types of word-of-mouth communication have emerged as a result of
technological innovations. We briefly summarize key concepts and position MWOM within the
word-of-mouth literature.
6
The Role of TWOM and EWOM in Consumer Decision Making
Building on initial work on opinion leaders by communications scholars Katz and Lazarsfeld
(1955), Arndt (1967) offered a first discussion of the defining characteristics of word of mouth.
To distinguish this original type of word-of-mouth communication from later types, we refer to it
as traditional word of mouth (TWOM). Arndt (1967) describes TWOM as face-to-face
communication about a commercial entity or offering between consumers, emphasizing its
unbiased and personal character. Although he does not explicitly distinguish between evaluative
post-purchase communication and anticipatory pre-purchase communication, ensuing TWOM
research has focused mainly on the former. Early TWOM research stressed the role of positive
word of mouth (i.e., recommendations or referrals) for product adoption decisions (Bass 1969;
Dodson and Muller 1978); interest in negative word of mouth only arose in relation to the
consumer satisfaction concept in the 1980s, and was considered a consequence of a consumer’s
dissatisfaction with a product (Richins 1983; Singh 1990). In addition to word of mouth as a
diffusion parameter, TWOM research has also dealt with a consumer’s decision to spread
positive or negative word of mouth (de Matos and Rossi 2008), with TWOM as the dependent
variable (Anderson 1998).
The rise of the Internet then enabled a new type of word-of-mouth communication: EWOM
(Godes and Mayzlin 2004). EWOM refers to online posts from mostly anonymous consumers
regarding a commercial entity or offering, that are visible to potentially millions of consumers,
and available for an indefinite period of time (Hennig-Thurau et al. 2004). However, EWOM
lacks the interpersonal connection between the sender and the receiver that is typical of TWOM,
which reduces its persuasive impact (Chatterjee 2001). As a result of EWOM’s observable
nature, scholars can analyze actual messages that consumers post, which has led to a
differentiation between message sentiment (i.e., valence) and the amount of messages (i.e.,
7
volume). Findings regarding the role of the valence dimension in driving sales are conflicting
though: Chevalier and Mayzlin (2006) and Chintagunta, Gopinath, and Venkataraman (2010)
report that EWOM valence affects product sales, whereas Liu (2006) finds no relation to success.
The volume dimension of EWOM does not carry any quality information but rather captures the
“buzz” (i.e., awareness and interest) that a commercial offering generates among consumers (Ho,
Dhar, and Weinberg 2009). Most research has found an influence of EWOM volume on product
adoption and sales (e.g., Godes and Mayzlin 2004 and Liu 2006), though Chintagunta, Gopinath,
and Venkataraman (2010) do not.
MWOM as New Type of Word-of-Mouth Communication
Microblogging word of mouth (MWOM) combines key elements of TWOM and EWOM that
are essential for their respective effectiveness. We define MWOM as any brief statement made
by a consumer about a commercial entity or offering that is broadcast to some or all members of
the sender’s social network through a specific web-based service (e.g., Twitter).2 Because
MWOM is sent electronically to a potentially very large network of personal connections, it
could reach audiences of similar size as those available through EWOM. Moreover, receivers are
directly connected to the sender and should be similarly susceptible to the personal influence that
characterizes TWOM messages. Also similar to TWOM, MWOM reaches network participants
in real-time, often sent from smartphones or other mobile devices. It is this real-time character in
combination with the large network of receivers that allows MWOM to spread more rapidly than
any other type of word-of-mouth communication, raising the question whether MWOM affects
product adoption and success earlier in a product’s lifecycle, when neither TWOM nor EWOM
2 We exclude messages sent to a single recipient through web-based services, as there is no conceptual difference between such communication and TWOM (which also includes mediated exchanges, such as phone calls and letters). See Hennig-Thurau and colleagues (2004) for a similar argument for EWOM.
8
can yet exert an impact. With Figure 1 we illustrate how MWOM combines characteristics of
both TWOM and EWOM and thus emerges as a new type of word-of-mouth communication.
--------------Figure 1 approx. here--------------
Empirical research on MWOM is still in an embryonic stage. Jansen and colleagues (2009)
collect data from Twitter but limit their insights to descriptive findings. For 24 movies, Asur and
Huberman (2010) use the rate of tweets to predict opening weekend box office, and combine this
measure with the tweets’ valence to predict the subsequent weekend box office. However, they
do not control for other types of word-of-mouth communication (e.g., EWOM) or producer
signals (e.g., advertising) when studying subsequent success, so while their findings might shed
light on the predictive potential of MWOM, they do not reveal whether MWOM causally affects
early adoption and success. To date, no published study has tested the “Twitter effect”—that is,
the impact of quality-related consumer information spread through MWOM messages on early
product adoption.
CONCEPTUAL MODEL AND HYPOTHESES
We summarize our conceptual model and hypotheses in Figure 2. We argue that the quality
judgments of consumers, articulated and spread immediately after the release of a new product
through MWOM, should influence subsequent early product adoption. This impact constitutes
the “Twitter effect.” We posit that two boundary conditions moderate the strength of this
“Twitter effect,” namely, the volume of evaluative MWOM articulated immediately after the
release of a new product and the differing susceptibility of different consumer segments to
MWOM.
--------------Figure 2 approx. here--------------
9
The “Twitter effect” has particular economic relevance for industries characterized by short
lifecycles and exponentially decaying adoption patterns, such as experiential media industries
(e.g., movies, music, electronic games), as those industries’ business models rely heavily on
large-scale early adoption. For example, approximately 50% of album sales of hit music (Asai
2009), 46% of movie ticket sales for major movies (Hayes 2002), and 40% of game revenues
(www.vgchartz.com) are generated in the first week of release. Before the advent of MWOM,
consumers faced information asymmetry and had to make early adoption decisions for
experiential media products on the basis of producer-provided quality signals only (e.g.,
advertising; Kirmani and Rao 2000), as very limited quality judgment of other consumers were
available and professional reviews are usually of limited informational value for consumers
(Eliashberg and Shugan 1997).
We argue that it is at this point that MWOM fundamentally challenges the status quo. Its
unique combination of characteristics expedite the spread of evaluative messages from
consumers who have experienced the new product, so quality judgments articulated by
consumers through MWOM can affect others’ product adoptions much earlier in the product’s
life. With MWOM, consumers share their quality evaluations with a vast network of followers
immediately after or even while consuming the new product in question; many consumers tweet
about the quality of a movie while they are still sitting in the theater. In turn, the valence of this
evaluative MWOM (hereafter, MWOM valence) should influence other consumers’ early
adoption decisions and thus the new product’s success. Valence, a concept adapted from emotion
research, describes the positive or negative emotional tone of MWOM, which is based on the
sender’s consumption of the new product (Brosch and Moors 2009).
10
If consumers who have experienced the new product spread mostly positive MWOM
messages immediately after its release, the product’s adoption in subsequent hours and days
should increase. If post-consumption MWOM messages are negative though, adoption should be
adversely affected. In other words, we propose the “Twitter effect” to be positive, as expressed in
our first hypothesis:
H1: The valence of MWOM messages spread by consumers who have experienced a new
experiential media product immediately after its release (“MWOM valence”) has a
positive effect on the product’s subsequent early adoption.
Beyond this direct effect, we predict a number of moderators. The first construct that we
argue to moderate the “Twitter effect” is MWOM volume, which we define in the context of this
research as the number of evaluative MWOM messages about a product spread immediately
after its release. The postulated impact of MWOM valence is based on the assumption that
positive or negative information about a new product reaches a large group of consumers, who
then, based on the information provided, make or change their adoption decisions. However, the
MWOM valence construct itself does not contain information about the number of consumers
reached by MWOM. We thus turn to the EWOM literature that has introduced the concept of
EWOM volume as a measure of the number of messages spread (Godes and Mayzlin 2004; Liu
2006), but restrict our MWOM volume construct to those messages sent by consumers who
already have experienced the new product and then spread positive or negative information about
its quality, as our interest is in the spread of such evaluative information (in contrast to mere
expressions of anticipation).
11
The interaction between MWOM valence and MWOM volume, articulated immediately after
a new product’s release, should capture the reach of evaluative MWOM messages. This reach in
turn should moderate the “Twitter effect.” Specifically, we expect a high MWOM volume to
amplify the impact of quality information contained in evaluative MWOM messages, whereas a
low volume should reduce its impact. We summarize this argument in our second hypothesis:
H2: The impact of the valence of MWOM messages spread by consumers who have
experienced a new experiential media product immediately after its release (“MWOM
valence”) on the product’s subsequent early adoption varies with the overall volume of
these MWOM messages (“MWOM volume”).
Moreover, we propose that consumer segments vary in their susceptibility to MWOM
influences, which should moderate the “Twitter effect” as a second boundary condition. We
focus on two consumer segments that are nowadays widely considered crucial for the success of
experiential media products, namely, teenagers and families (e.g., Epstein 2010; Fromme 2003).
We are not aware of any research on consumers’ susceptibility to MWOM influence; however,
existing consumer research has demonstrated that consumers vary in their susceptibility to
reference group influences in general (Childers and Rao 1992). Existing consumer research has
shown that teenagers are particularly susceptible to peer influence in the context of shopping
decisions, and that they enjoy their shopping experiences more when they are shared with and
validated by respected peers (Bachmann, Roedder John, and Rao 1993; Mangleburg, Doney, and
Bristol 2004). Building on these findings, we expect that teenagers’ purchase decisions are more
strongly influenced by MWOM messages than are those of average consumers and that the
relationship between MWOM valence and product adoption is stronger for these consumers. As
12
a result, the early adoption of experiential media products whose core target group consists of
teenagers therefore should be more strongly affected by the “Twitter effect” than should products
that focus on other target groups.
Families, as a second key target group for experiential media products, are decision-making
entities that must negotiate joint decisions across individual members (Commuri and Gentry
2000). Although parents are the dominant decision-makers, children exert considerable influence
on particular sub-decisions, such as leisure activities like vacations, movie attendance, and
television viewing (Jenkins 1979; Mangleburg 1990). No research, to the best of our knowledge,
has explicitly examined families’ susceptibility to outside influences such as MWOM, but the
negotiated nature of entertainment-related consumption choices suggests that family decisions
should be less affected by MWOM valence than those of average consumers. Consistent with
this argument, recent research on family identity and its role in family decision making has
highlighted the importance of leisure activities as opportunities for family identity enactment
through collective experiences (Epp and Price 2008). The centrality of collective experiences to
family identity implies that the overall joint consumption experience is only partly influenced by
the (anticipated) quality of a new experiential media product, so that valenced information about
product quality transmitted by MWOM should play a lesser role in family decisions.
Consequently, we expect the early adoption of experiential media products whose core target
group are families to be less strongly affected by the “Twitter effect” than the early adoption of
products that focus on other target groups. We offer our third and final hypothesis:
H3: The impact of the valence of MWOM messages spread by consumers who have
experienced a new experiential media product immediately after its release (“MWOM
valence”) on the product’s subsequent early adoption varies for different consumer
13
segments according to their susceptibility to MWOM influence, such that the impact is
(a) stronger for experiential media products targeted at teenage consumers, and (b)
weaker for experiential media products targeted at families.
EMPIRICAL STUDY: TESTING THE “TWITTER EFFECT”
Context, Research Design, and Sample
To test our hypotheses, we collected data in the context of the motion picture industry. We
chose movies because (a) they are an economically important category of experiential media
products, (b) the debate about the existence of the “Twitter effect” is prominent for movies, and
(c) we were able to compile data on daily revenues and important controls (e.g., advertising
spending) for major new releases. Since we are interested in early product adoption, we focused
on movies’ opening weekend. Early adoption is critical for movie success and accounts for
approximately 46% of total movie ticket sales (Hayes 2002); research provides evidence of
additional effects on future consumer adoption decisions and distribution choices (e.g., Elberse
and Eliashberg 2003). In North America, movies are generally released in theaters on Fridays.
We thus use the valence of MWOM messages sent within the first 24 hours after a movie’s
release on Friday by consumers who viewed the movie and study its impact on the movie’s
North American theatrical box office revenues for the remainder of the weekend (i.e., Saturday
and Sunday). In other words, Saturday and Sunday box office revenue is our measure of
subsequent early new product adoption. As both TWOM and EWOM for new movies require
more time to spread on a large scale (even popular EWOM sites such as the Internet Movie
Database IMDb generally do not report consumer opinions before Monday), this research design
allows us to isolate the potential impact of MWOM.
14
We collected data on all movie titles that were widely released in North American theaters
(i.e., shown simultaneously in more than 800 theaters at their release) between October 2009 and
October 2010; we excluded 11 titles that were released on different days of the week to avoid
any possible bias.3 The final sample consists of 105 movie titles (see the Appendix for a
complete list).
Empirical Model
The dependent variable in the empirical model that we developed to test our hypotheses was
the total North American box office revenue generated by a movie on the Saturday and Sunday
of its release weekend. As independent variables, we included the MWOM valence of evaluative,
post-purchase messages sent within the first 24 hours after a movie’s Friday release; the
interaction of MWOM valence and the overall volume of these evaluative MWOM messages;
and the interactions of MWOM valence with a movie’s target groups of teenagers and families.
We also included a number of control variables derived from extant motion picture research to
rule out alternative explanations and confounding effects.
Specifically, we included the release day (i.e., Friday) revenues as a control, so the model
only focuses on that part of the Saturday and Sunday box office that is not accounted for by the
general appeal of the movie, which already has been reflected in its release day success.
Therefore, our model targets the variation of Saturday and Sunday box office from the release
day success, not the absolute success of the movie. With this model specification, we ensured
Granger (1969) causality when testing for the “Twitter effect.”
We also controlled for potential influences of established determinants of motion picture
success, namely, a movie’s starpower (Elberse 2007), whether it is a sequel (Hennig-Thurau,
3 In addition, technical problems caused by the Twitter API meant 12 movies could not be considered in the analyses.
15
Houston, and Heitjans 2009), the production budget (Basuroy, Chatterjee, and Ravid 2003),
advertising spending (Elberse and Eliashberg 2003), and its pre-release buzz (Karniouchina
2011). We also included professional critics’ ratings (Basuroy, Chatterjee, and Ravid 2003), the
only independent quality judgment available upon a movie’s release. To estimate the interaction
effects, we also included the respective main effects, namely, evaluative, post-purchase MWOM
volume and the movie’s target groups (teenagers and families).
In a separate equation, we used release day revenues as the dependent variable and the
aforementioned success factors as independent variables. We did so to account for the
established effects of these variables on release day success. Because the release day success
variable is also included in the Saturday and Sunday box office revenues equation as an
independent variable for the reasons mentioned above, this equation helped us avoid model
misspecifications.
We followed Elberse and Eliashberg (2003) and chose a multiplicative log-linear model
formulation, as shown in Equations 1 and 2:
(1)
0 31 2
5 64
7 8 9
( ) ( )( ) MWOMVAL AUDFAM MWOMVAL AUDTEENS
D ME
REV_SUB e REV_REL MWOMVAL MWOMVOL
MWOMVAL MWOMVOL e e
X e e e
(2) 0 1 2ß DREV_REL e X e e
REV_SUB is the North American theatrical box office revenues generated by a movie during
the Saturday and Sunday after the movie’s release; REV_REL is the North American theatrical
box office revenues generated by a movie on its release day (i.e., Friday); MWOMVAL is our
measure of MWOM valence; and MWOMVOL is our measure of MWOM volume. AUDFAM
and AUDTEENS are dummy variables that indicate whether a movie’s main target group is
families or teenagers, respectively; X is a vector of metric control variables (i.e., a movie’s
16
budget, amount of advertising spent for the movie before its release, its pre-release buzz, and its
reception by professional critics); D is a vector of dummy-coded control variables (i.e., the
movie’s starpower and whether it is a sequel to a previous film); and ME is a vector that consists
of the main effects of the two target group variables.
In Table 1 we provide a description of all variables in our empirical model, their
operationalization, and their empirical and intellectual sources. For the key concepts of MWOM
valence and the moderators (i.e., MWOM volume and the two customer segments, teenagers and
family), we provide additional details about their operationalization below.
--------------Table1 approx. here--------------
MWOM valence. For the 105 movies in the sample, we collected all English-language
MWOM messages sent via Twitter on each day of the opening weekend. We used Twitter
messages as a proxy for MWOM messages in general for two reasons. First, Twitter is by far the
largest microblogging platform (Knowlton 2011), regularly being used as a synonym for
microblogging in general (Anamika 2009). Second, Twitter enabled us to download all tweets
sent about a movie during the opening weekend in real time by granting us extended access
rights to their Application Programming Interface (API). This access was essential, because for
major movies, the amount of Twitter chatter often drastically exceeds the API’s normal
download limits. No existing studies on microblogging (e.g., Asur and Huberman 2010; Jansen
et al. 2009) report a similar rights extension.
Every week from October 2009 to October 2010, we developed a list of search terms for each
movie due to be released on the Friday of that respective week. One author generated an initial
set of search terms based on an extensive manual Twitter search, which was then reviewed and
discussed among all authors to ensure completeness. Up to ten search term combinations were
17
considered per movie, taking into account Twitter-specific acronyms, exclusion words, and so
on. These search term combinations then were manually entered into a script, which
automatically downloaded all tweets containing the specified search term combinations
throughout the opening weekend, starting on Fridays at 10:00 a.m. Eastern Daylight Time and
ending Sunday at midnight Eastern Daylight Time. Overall, we collected 4,045,350 tweets about
the 105 movies in our sample. Our extended access rights ensured that these tweets include all
English-language MWOM messages spread about the movies in our sample through Twitter.
Although it is not possible to collect information about the number of followers per tweet due to
Twitter’s privacy policy, the Max Planck Institute (2011) estimated that the average number of
followers per Twitter user is approximately 45, which suggests that the tweets in our sample
have reached roughly 182 million consumers total.
Using this information, we operationalized MWOM valence as the quotient of all positive
tweets a movie received on its opening Friday and the following Saturday until noon, divided by
the number of negative tweets for the movie within the same time period. To determine the
valence of the individual tweets, we ran a multistage sentiment analysis. Prior to the actual
analysis, we eliminated all tweets with identical content by the same author and those not written
in Latin script. The sentiment analysis then involved two steps. First, all remaining tweets were
classified into one of three groups: (1) spam, non-English tweets, and tweets not related to the
movie in question, (2) movie-related tweets that contained no post-consumption quality
assessment (mostly anticipatory statements such as “I look forward watching MOVIE A
tonight”), and (3) evaluative, post-consumption tweets (movie “reviews”). Second, we divided
the third group into positively and negatively valenced tweets using sentiment analysis.
18
The analysis was executed simultaneously for all movies, employing the open-source data
mining software WEKA (Bouckaert et al. 2010; Hall et al. 2009). Initially we manually coded
51,000 randomly selected tweets into the three aforementioned categories. Using 65% (i.e.,
33,150) of these coded messages as input, we trained the algorithm of a support-vector machine
(SVM) to build a model to classify cases into categories. Through a decomposition of the
manually coded tweets into their elements (i.e., single words and word groups), these elements
were used to calibrate the model, identifying each element’s discriminatory power. More
formally, a vector was assigned to all words and word groups and mapped into a multi-
dimensional space. The SVM then fitted a hyperplane that divides all training points (i.e.,
vectors) into two classes, such that it maximized the distances between the hyperplane and the
nearest training points. Then the SVM identified those words and word groups whose vectors
showed the greatest distance from the hyperplane and assigned a parameter to each, indicating
the strength of association with a particular category. The words and word groups with the
highest discriminatory power were used for further analysis (Pang, Lee, and Vaithyanathan
2002).
We next applied the SVM to classify all other (non-coded) tweets. Using the sequential–
minimal–optimization algorithm, the SVM searched for the previously identified words and
word groups in each of these tweets. The previously determined parameters of the recognized
words and word groups were then used to calculate the degree to which each tweet was
associated with the different categories, resulting in the final classification of all collected tweets
(Keerthi et al. 2001; Platt 1999). To determine the predictive power of this classification
analysis, we ran an out-of-sample test with the remaining 35% (i.e., 17,850) of the manually
coded tweets that were not used to calibrate the model. These tweets were classified as positive
19
and negative reviews, with an accuracy level of 90.2%—higher than most other studies that use
sentiment analysis to code online consumer articulations on the web (e.g., Das and Chen 2007),
which may be a result of the brevity of MWOM messages compared with EWOM messages.
Moderator variables. Drawing on the same data and classification results we used to measure
MWOM valence, we operationalized MWOM volume as the number of the evaluative, post-
consumption tweets for a movie (movie “reviews”) sent on a movie’s opening Friday till the
following Saturday until noon. For the customer segment moderators, we classified each movie
according to three nominally scaled variables: “teenagers as main target group,” “family as main
target group,” and “other main target group.” This classification relied on Jinni.com, a site on
which experts assign movies to audience groups. For this research, we reduced an original six-
group classification to three groups by merging Jinni’s “family outing” and “kids” categories,
using its “teens” category, and then including all movies not assigned to one of these groups in
the “other main target group” category.
To create the interaction terms between MWOM valence and MWOM volume, we used
Lance’s (1988) residual centering approach to minimize the potential multicollinearity between
the interaction term and the main effects (Bottomley and Holden 2001; Hennig-Thurau, Houston,
and Heitjans 2009). Residual centering is an effective, conservative test for interaction effects of
metric variables; it assigns only that part of the variance to the interaction term that is not
explained by the main effects and does not suffer from problems associated with mean-centering
(Echambadi and Hess 2007). We used the raw product terms for the interactions between
MWOM valence and the two target group variables, as the latter are dummies which limits the
usefulness of residual centering.
20
Estimation
To estimate the system of Equations 1 and 2, we first linearized these equations, which
resulted in Equations 3 and 4:
(3) 0 1 2 3
4 5
6 7 8 9
( ) ( ) ( ) ( )
( ) ( )
( ) ( )
LN REV_SUB ß ß LN REV_REL ß LN MWOMVAL ß LN MWOMVOL
ß LN MWOMVAL MWOMVOL ß MWOMVAL AUDFAM
ß MWOMVAL AUDTEENS ß LN X ß D ß ME
(4) 0 1 2( ) ( )LN REV_REL ß ß LN X ß D
We then estimated Equations 3 and 4 using seemingly unrelated regression (SUR; Zellner 1962),
which provided unbiased coefficients for the model variables by accounting for correlated errors
across the two equations stemming from the dual role of REV_REL as both an independent
variable in Equation 3 and a dependent variable in Equation 4. For SUR to be effective, the
system of equations must contain at least one regressor that is used in one equation but not the
other (Elberse 2010). In our model, MWOM valence, the moderator main effects, and the
corresponding interaction terms all meet this criterion and are included in Equation 3 but not
Equation 4.
Results
Descriptive statistics. Of the approximately four million tweets we collected, 829,576 were
classified as evaluative, post-consumption MWOM. The number of review tweets per movie
varied; the mean is 38,527. Consistent with previous studies on MWOM (e.g., Asur and
Huberman 2010) and also EWOM (e.g., Chevalier and Mayzlin 2006), the reviews were more
positive than negative. Figure 3 depicts the number of movie-related tweets sent throughout the
opening weekend, differentiating between movie “reviews” and pre-purchase movie-related
tweets that did not contain quality information. Friday was the most active day in terms of
spreading MWOM; approximately 51% of all movie-related tweets (2,072,731) and 65% of
21
movie “reviews” (543,836) were sent from Friday till the following Saturday until noon. Both
kinds of tweets peaked on the opening day and declined thereafter. Across all three days of the
weekend, review tweets peaked at around 11:00 p.m. Eastern Daylight Time, which is about
three hours after the peak of non-review tweets. This distribution appears consistent with our
prediction that most non-review tweets would be sent shortly before the screening (expressing
anticipation), whereas movie “reviews” are mostly sent after the screening (expressing
evaluation). Table 2 reports the basic descriptive statistics and correlations.
--------------Figure 3 and Table 2 approx. here--------------
Hypotheses tests. The overall fit of the model was good, with R-square values of .62 for the
REV_REL equation and .98 for the REV_SUB equation (which included the release day revenues
as a regressor). Multicollinearity was below critical levels and thus not a problem (Hair et al.
2006). The variance inflation factor (VIF) for MWOM valence, our key construct, was 2.6, and
the highest VIF of all regressors was 5.2 (for the interaction between MWOM valence and
teenagers as the main target group).
As we report in Table 3, the results of the SUR estimation provide support for H1 and thus for
the existence of the “Twitter effect.” The MWOM valence parameter in the REV_SUB equation
is positive and significant, consistent with our expectations: MWOM valence spread immediately
after a new movie’s release systematically influences other consumers’ decisions about whether
to attend a screening of the movie during the remainder of its opening weekend.
--------------Table 3 approx. here--------------
For the moderator hypotheses, we found mixed support. The SUR parameter for the
interaction between MWOM valence and MWOM volume is in the expected direction but not
significant (p = .105); we thus do not find support for H2. For H3, the interaction between
22
MWOM valence and teenagers as the main target group is also non-significant (p = .96).
However, the interaction between MWOM valence and families as the main target group is
significant and negative, as theoretically proposed (p < .01). The positive effect that MWOM
valence exerts on subsequent early product adoption is attenuated if families are the main target
group. Thus, though we do not find support for H3a, we do find support for H3b.
The parameters of the controls are generally consistent with extant theory. Anticipatory buzz
about a new movie has the strongest effect on release day revenues, followed by whether the
movie is based on a prominent brand (e.g., sequel) and then its production budget (as an
indicator of the movie’s production values). Critical reviews are only marginally significant,
while pre-release advertising and starpower are not. We might speculate that their effects are
accounted for by both the buzz a movie creates and its production budget.
Regarding Saturday and Sunday revenues, family-targeted movies show a positive main
effect, and the production budget also positively affects the remainder of the opening weekend.
The sequel variable, MWOM volume, buzz, and advertising all exhibit negative coefficients in
this equation (though non-significant for the latter two). All of these variables are skewed toward
release day revenues (compared with the remainder of the weekend) and are particularly high for
movies that drive large audiences into theaters on a movie’s opening night. Because we
controlled for release day revenues in Equations 1 and 3, their negative parameters in the latter
equation would indicate that movies that have strong brands and are accompanied by high
MWOM volume, buzz, and advertising tend to peak on their release day, when the fan base
crowds theaters, and then attract relatively smaller audiences on the days that follow. In other
words, their parameters in this system of equations must not be confused with causal effects.
23
FOLLOW-UP ANALYSIS OF THE ROLE OF QUALITY-RELATED INFORMATION FOR THE “TWITTER EFFECT”
Our analysis thus far has provided empirical support for the existence of the “Twitter effect,”
in that the valence of consumer judgment about a new experiential media product spread through
MWOM affects subsequent early adoption and new product success. A key assumption
embedded in this argument is that the information about a product’s quality, as perceived by
consumers, is responsible for this effect. To provide additional support for our main hypothesis
and rule out alternative explanations, we conducted a follow-up analysis in which we substituted
MWOM valence with a direct aggregate measure of consumers’ quality perceptions of a new
product, a concept referred to in the literature as ordinary evaluation (Holbrook 2005; Holbrook
and Addis 2007). We then compared the effect of this ordinary evaluation on subsequent early
adoption between a “MWOM period” sample and a “pre-MWOM period” sample of movies.
That is, we tested the effect of ordinary evaluation on the box office revenues generated by a
movie during the remainder of its initial weekend (i.e., Saturday and Sunday), when entered into
our early adoption equation (i.e., Equations 2 and 4) as a substitute for MWOM valence for the
two different samples.
The first sample contained the recently released films we used in our preceding analyses; it
represents the “MWOM period” sample. The “Twitter effect” would suggest that ordinary
evaluation has an effect in this sample, because MWOM enables the instant spread of post-
purchase evaluations. The second sample comprised similar movies that were released before
Twitter and other MWOM services became popular and thus before the spread of quality-
information through MWOM was possible for consumers; we thus refer to it as the “pre-MWOM
period” sample. To identify these “similar movies,” we drew on a comprehensive database of
1,202 movies theatrically released in North America between 1998 and 2006 on at least 800
24
screens during their release weekend. Twitter began operations in 2007, when it hosted just
5,000 tweets—a tiny fraction of today’s 190 million daily messages. Because no other means
existed to enable ordinary evaluation to spread as fast, we can assume that ordinary evaluations
should have no discernible effect on subsequent early adoptions in the pre-Twitter era.
As a proxy for ordinary evaluation, we collected each movie’s rating on the video rental site
Netflix, the leading North American provider of DVD-by-mail and VOD movie streaming, for
both the “MWOM period” and “pre-MWOM period” samples. These Netflix ratings reflect the
quality perceptions of its 23 million North American customers, a mainstream audience
(Mullaney 2006) that is consistent with the definition of ordinary evaluation as a measure of the
taste of “ordinary” consumers (i.e., non-experts or members of the mass audience; Holbrook
2005).
Matching Approach
To compare the role of ordinary evaluation on subsequent early adoption for two time periods
(i.e., “MWOM period” and “pre-MWOM period”), we needed to ensure that the samples did not
differ systematically, which might have distorted the results. For example, such differences
might arise from structural changes in the movie industry’s selection process, which currently
focuses more on so-called “tentpole” pictures (e.g., Avatar), and on franchises (e.g., Pirates of
the Caribbean), while producing fewer medium-budgeted, unbranded films.
To generate such a sample for the “pre-MWOM period,” we applied propensity score
matching, a statistical matching procedure (Rosenbaum and Rubin 1983), and used the
propensity scores for each of the movies generated by this approach to identify a “twin” for each
of the 105 movies in our “MWOM period” sample, using nearest neighbor estimation.
Propensity score matching has been developed to remove selection biases between treatment
groups and no-treatment groups in non-experimental settings (for applications in marketing, see
25
Mithas and Krishnan 2009 and von Wangenheim and Bayón 2007). In our case, the treatment is
time and the changes in the movie industry that accompany it. Consequently, the “MWOM
period” movies represent the treatment cases, and the “pre-MWOM period” movies are the no-
treatment cases.
Propensity score matching applies probit regression (with the “MWOM period” variable as
the dependent variable, coded 0 or 1 for each movie) to generate a propensity score for each
sample element, which then provides the basis for the subsequent steps. As the regressors, it uses
a set of variables that should differ systematically between the treatment (i.e., “MWOM period”)
and no-treatment (i.e., “pre-MWOM period”) cases and that also affect the outcome variable
(i.e., subsequent early adoption). Therefore, we used a set of movie characteristics that previous
research has found to influence box office success, as discussed in the context of Equations 1 and
2. We report the model and results of the probit regression estimation in Table 4. The regression
function was highly significant and, with a pseudo-R-square value of .22, able to explain
differences between the two groups.
--------------Table 4 approx. here--------------
We used the resulting propensity scores to identify “nearest neighbors” for each movie in the
“MWOM period” sample; the nearest neighbor from the “pre-MWOM period” sample was the
movie with the smallest Euclidean distance to an “MWOM period” movie’s propensity score.
We chose nearest neighbor estimation over alternative algorithms such as Kernel matching, as
the latter does not identify one twin for each sample unit but rather would use a weighting score,
which is inconsistent with our application of seemingly unrelated regression for the subsequent
early adoption equation.
26
As Table 5 shows, all mean differences between the treatment and no-treatment cases that
were significant before the matching process (i.e., budget, advertising, pre-release buzz, ordinary
evaluation, and teenagers as target audience) became insignificant after the matching. As another
proof of matching effectiveness, we find that when we reran the probit regression with the
matched sample, the pseudo-R-square value was substantially smaller after matching (.03) than
before. Thus, propensity score matching successfully removed differences between the “MWOM
period” and the “pre-MWOM period” samples, allowing us to compare the SUR coefficient for
the ordinary evaluation variable in the subsequent early adoption regression.
--------------Table 5 approx. here--------------
SUR Results for “MWOM Period” and Matched “Pre-MWOM Period” Samples
In the final step, we estimated our system of equations again using SUR for both samples, so
that we could determine if ordinary evaluation had a significant effect on box office revenue in
either time period. In other words, we tested if the ability to spread quality-related information
has changed over time and can influence subsequent early adoption of movies today, which
would be additional evidence of the proposed “Twitter effect.” We ran SUR separately for the
“MWOM period” and “pre-MWOM period” samples, essentially replicating Equations 3 and 4.
The only changes were that we substituted MWOM valence with ordinary evaluation and
excluded the valence-based interactions. We report the results in Table 6.
--------------Table 6 approx. here--------------
Consistent with our theoretical arguments, ordinary evaluation had no significant effect on
subsequent early adoption in the “pre-MWOM period” sample (p = .99). In contrast (but in line
with our arguments), in the “MWOM period” sample, we find that ordinary evaluation exerts a
significant positive effect on subsequent early adoption (p < .05). Thus, our matched sample
follow-up analysis provides evidence that the “Twitter effect” of MWOM valence indeed results
27
from MWOM’s power to spread consumers’ ordinary evaluation (i.e., their post-purchase quality
evaluations of a new product) rapidly among large groups of relevant consumers, which then
affects the new product’s early adoption.
DISCUSSION, IMPLICATIONS, AND RESEARCH OPPORTUNITIES
Summary
This research introduces the concept of MWOM to the marketing literature, extending
previous research on TWOM and EWOM. MWOM differs from these concepts in that it
combines the personal influence element of TWOM with EWOM’s ability to reach very large
audiences, while at the same time drastically increasing the speed of information dissemination.
We argue that MWOM challenges existing business models adopted by producers of experiential
media products, because it reduces the information asymmetry typical for these products, whose
adoption follows an exponential decay pattern and whose initial successes are amplified through
action-based information cascades.
We developed a conceptual model of this “Twitter effect,” including boundary conditions in
the form of moderator effects, and tested the proposed effect using all MWOM messages sent
through Twitter during the opening weekends of 105 movies widely released in North American
theaters between October 2009 and 2010. Our findings support the existence of the “Twitter
effect,” the intensity of which varies with a movie’s target group, such that it is weaker for
family films. Our follow-up analysis with a matched sample of 105 movies released before
MWOM became a mass phenomenon (i.e., between 1998 and 2006) suggests that the “Twitter
effect” is indeed the result of the quick spread of consumers’ post-purchase quality evaluations
of these movies, as enabled by MWOM.
28
Managerial Implications
These findings have substantial implications for marketing managers, particularly those who
are responsible for the success of experiential media products. Most significantly, our findings
demonstrate that the information asymmetry between producers and consumers which
traditionally exists at the release of a major experiential media product is indeed reduced by the
availability of MWOM. We provide evidence that for motion pictures, MWOM helps spread
evaluative post-purchase quality opinions about experiential media products so quickly and
widely that it significantly affects subsequent adoption behaviors already on the next day.
As a result of the rise of MWOM, managers responsible for such products have less of a
“buffer” to insulate them from consumer opinion. Consider the disappointing opening weekend
box office results for the movie Brüno, whose sales plummeted 40% from Friday to Saturday and
lost even more momentum going into Sunday, supposedly due to negative MWOM spread on
Twitter (Van Grove 2009). The reduced information asymmetry between producers and
consumers offers both a chance and a threat to producers; it is mainly a threat to those products
that consumers perceive as low in quality. Our findings thus should motivate producers to
increase their focus on developing high-quality new products that really meet consumer needs,
and then marketing them in a way that truthfully reflects their quality. Such products will benefit
from reduced information asymmetry, because MWOM spreads good news about their quality
quickly among networked consumers.
But the “Twitter effect” also carries more fundamental implications. Because experiential
media product quality results from a complex creative process, producing only high-quality
products is virtually impossible. The “blockbuster” business model that dominates these
industries requires information asymmetry at and shortly after the new product’s release, so that
producers can redeem investments in products although they have turned out to be creative
29
failures. Actually, (movie) producers have systematically transformed their distribution and
production pattern, from one that builds on word-of-mouth communication to one that relies on
mass advertising and mass distribution and exploits information asymmetries (for a review of
this transformation process, see PBS 2001), and other experiential media industries have
followed suit. Because post-purchase, evaluative word of mouth could not spread quickly and
widely enough before the advent of MWOM, this “blockbuster business model” virtually
guaranteed release success, at least for products that were deemed interesting enough to
stimulate strong buzz, mainly in response to heavy advertising. To address the existence of
MWOM and its impact on early adoption, producers will need to adjust their business model to
the changed environment. For example, they might return to a more word-of-mouth dependent
business model, though such an option conflicts with the high budgets allocated to the
production and marketing of blockbuster titles. No company can afford to invest $250 million in
a movie and only show it on very few screens (and write off the investment if TWOM, EWOM,
and MWOM are negative). Such a business model would thus also imply the descent of big
budgets and blockbusters.
An alternative reaction, consistent with the quality imperative, would be to retain the current
focus on brand names and franchises but base decisions about which brands to turn into a movie
more on consumer-perceived quality. This strategy would acknowledge the higher risk of
producing experiential media products in the MWOM era but also allow producers to invest in
blockbusters (and enjoy the advantages of such products in a globalized world).
Finally, producers might try to influence MWOM directly through MWOM marketing or
social media efforts, such as the recommender programs on Facebook or Twitter, which could
might attenuate negative messages shared by other consumers. Although such a strategy seems
30
currently favored by media industries (Singh 2009), it remains unclear whether industry-
produced information can actually influence MWOM valence (compared with buzz and MWOM
volume). It also is imperative for experiential media product managers to devise a
communications strategy that monitors and tries to carefully steer MWOM communication
during the opening weekend, especially on the release day (Moviemarketingmadness 2009).
Such a communications strategy is particularly important for movies targeted at teenagers, who
are, as we find, comparatively more influenced by MWOM messages than are family audiences.
Research Implications
For researchers interested in word-of-mouth communication, it is important to recognize
MWOM as a distinct type of word-of-mouth that is characterized by the unique combination of
TWOM’s immediacy and personal influence and EWOM’s potential to reach large audiences. As
we have demonstrated, the resulting speed of information dissemination has a profound impact
on the adoption of certain products within as little as 24 hours. We therefore recommend
considering MWOM separately to fully understand word-of-mouth influences on product
adoption. How does MWOM interact with other types of word of mouth, such as TWOM and
EWOM, and how and to what extent do they all converge to produce a consensus judgment?
Very limited research has yet focused on the differences and similarities between the different
types of word-of-mouth communication. Such studies would be welcomed to better understand
the different types in general and their respective roles for consumer decision making in
particular (e.g., Berger 2011).
Previous research studying the impact of EWOM on product adoption suggests that especially
“buzz” (often equated with anticipatory, pre-purchase EWOM) influences adoption (Asur and
Huberman 2010; Liu 2006). But MWOM changes these dynamics, because post-purchase
evaluations can be disseminated very quickly and thus affect a product’s lifecycle much earlier
31
than has been previously possible. The focus of studies on personal influence therefore should
shift back from volume to valence in the MWOM era.
Our empirical research focuses on movies and Twitter, but it would be interesting to replicate
our results by studying the impact of MWOM spread through other channels, such as Facebook
status updates, on other experiential media products, such as computer games or books. We
provide arguments that the “Twitter effect” is not industry specific, but further research that
investigates the role of product context for this effect would be desirable.
Furthermore, this research adds to a stream of studies that employ secondary, aggregate-level
data, and we know as yet little about the effect of MWOM messages on individual consumers.
We infer such effects from the adoption behavior of the target audience, but we do not study
teenagers’ or families’ decision processes directly. Focusing on individual consumers’
motivations might also help reveal why there were more positive than negative MWOM
messages overall in our sample.
Whereas existing research has established a negativity bias in consumers, meaning that
consumers are more strongly influenced by negative rather than positive word-of-mouth
messages, recent research points to the possibility that temporal contiguity cues mitigate this
bias. For example, Chen and Lurie (2011) find that EWOM messages that refer to recent
consumption experiences are perceived as less useful. Because MWOM is characterized by its
immediacy, such message framing effects may be insightful and should be explored further.
CONCLUSION
This research introduces MWOM to the marketing literature as a new type of word-mouth
communication. Combining characteristics of TWOM and EWOM, MWOM influences
consumers’ subsequent early adoption of new movies by enabling consumers to spread their
32
post-purchase quality perceptions on large scale and very fast. This “Twitter effect” threatens
existing business models for experiential media and other industries, because it increases the
relevance of product quality for economic success and shrinks the window in which consumers
will adopt a new product without being able to rely on other consumers’ quality judgments.
33
REFERENCES
Anamika S. (2009), “Microblogging Sites–40 Twitter like Websites List,” (accessed January 19, 2012), [available at http://anamikas.hubpages.com/hub/40-Microblogging-Sites-list-for-Communication-Twitter-Alternatives].
Anderson, Eugene (1998), “Customer Satisfaction and Word of Mouth,” Journal of Service Research, 1(1), 5-17.
Arndt, Johan (1967), “Role of Product-Related Conversations in the Diffusion of a New Product,” Journal of Marketing Research, 4(3), 291-295.
Asai, Sumiko (2009), “Sales Patterns of Hit Music in Japan,” Journal of Media Economics, 22, 81-101.
Asur, Sitaram and Bernardo A. Huberman (2010), "Predicting the Future with Social Media," Proceedings of the 2010 IEEE/WIC/ACM International Conference on Web Intelligence and Intelligent Agent Technology, 1, 492-99.
Bachmann, Gwen R., Deborah Roedder John, and Akshay R. Rao (1993), "Children’s Susceptibility to Peer Group Purchase Influence: An Exploratory Investigation,” in Advances in Consumer Research, Vol. 20, Leigh McAlister and Michael L. Rothschild, eds. Provo, UT: Association for Consumer Research, 463-68.
Bass, Frank M. (1969), “A New Product Growth for Model Consumer Durables,” Management Science, 15 (January), 215-227.
Basuroy, Suman, Subimal Chatterjee, and S. Abraham Ravid (2003), “How Critical Are Critical Reviews? The Box Office Effects of Film Critics, Star Power, and Budgets,” Journal of Marketing, 67(4), 103-117.
Berger, Jonah (2011), “Different Drivers of online and Offline Word of Mouth,” Advances in Consumer Research, Proceedings of the 39th Annual Association for Consumer Research Conference, October 13-16, 2011, Saint Louis, MO, USA.
Bikhchandani, Sushil, David Hirshleifer, and Ivo Welch (1992), “A Theory of Fads, Fashion, Custom, and Cultural Change as Informational Cascades,” Journal of Political Economy, 100(5), 992-1026.
Bottomley, Paul A. and Stephen J. S. Holden (2001), “Do We Really Know How Consumers Evaluate Brand Extensions? Empirical Generalizations Based on Secondary Analysis of Eight Studies,” Journal of Marketing Research, 38, 494-500.
Bouckaert, Remco, Raymond Hemmecke, Silvia Lindner, and Milan Studený (2010), “Efficient Algorithms for Conditional Independence Inference,” Journal of Machine Learning Research, 11, 3453-3479.
34
Brosch, Tobias and Agnes Moors (2009), “Valence,” in The Oxford Companion to Emotion and the Affective Sciences, David Sander and Klaus R. Scherer, eds. Oxford: Oxford University Press.
Chatterjee, Patrali (2001), "Online Reviews: Do Consumers Use Them?” in Advances in Consumer Research, Vol. 28, Mary C. Gilly and Joan Meyers-Levy, eds. Valdosta, GA, Association for Consumer Research, 129-133.
Chen, Zoey, and Nicholas Lurie (2011), “Temporal Contiguity and the Negativity Bias in Online Reviews,” Advances in Consumer Research, Proceedings of the 39th Annual Association for Consumer Research Conference, October 13-16, 2011, Saint Louis, MO, USA.
Chevalier, Judy and Dina Mayzlin (2006), “The Effect of Word of Mouth on Sales: Online Book Reviews,” Journal of Marketing Research, 43 (August), 345-54.
Childers, Terry L. and Akshay R. Rao (1992), “The Influence of Familial and Peer-Based Reference Groups on Consumer Decisions,” Journal of Consumer Research, 19 (September), 198-211.
Chintagunta, Pradeep K., Shyam Gopinath, and Sriram Venkataraman (2010), “The Effects of Online User Reviews on Movie Box Office Performance: Accounting for Sequential Rollout and Aggregation across Local Markets,” Marketing Science, 29(5), 944–957.
Commuri, Suraj and James W. Gentry (2000), “Opportunities for Family Research in Marketing,” Academy of Marketing Science Review, 2000 (8), (accessed October 12, 2011), [available at http://www.amsreview.org/articles/commuri08-2000.pdf].
Corliss, Richard (2009), “'Bruno': Did Twitter Reviews Hurt Movie at Box Office?”(accessed August 17, 2010), [available at http://www.time.com/time/arts/article/0,8599,1910059,00.html].
De Matos, Celso Augusto and Carlos Alberto Vargas Rossi (2008), “Word-of-Mouth Communications in Marketing: A Meta-Analytic Review of the Antecedents and Moderators,” Journal of the Academy of Marketing Science, 36(4), 578-596,
Das, Sanjiv R. and Mike Y. Chen (2007),“Yahoo! for Amazon: Sentiment Extraction from Small Talk on the Web,” Management Science,53(9), 1375-1388.
Dodson, Joe A. and Eitan Muller (1978), “Models of New Product Diffusion Through Advertising and Word-of-Mouth,” Management Science, 24 (15), 1568-78.
Dugan, Lauren (2010), “Study: There Is no "Twitter Effect" Happening in Hollywood,” (accessed October 12, 2011), [available at http://socialtimes.com/study-there-is-no-twitter-effect-happening-in-hollywood_b23714].
Echambadi, Raj and Hess, James(2007), “Mean-Centering Does Not Alleviate Collinearity Problems in Moderated Multiple Regression Models,” Marketing Science, 26(3), 438-445.
35
Elberse, Anita (2007), "The Power of Stars: Do Star Actors Drive the Success of Movies?" Journal of Marketing, 71(4), 102-120.
——— (2010), "Bye Bye Bundles: The Unbundling of Music in Digital Channels." Journal of Marketing, 74 (3), 107-23.
——— and Jehoshua Eliashberg (2003), “Demand and Supply Dynamics for Sequentially Released Products in International Markets: The Case of Motion Pictures,” Marketing Science, 22(3), 329-354.
Eliashberg, Jehoshua and Steven M. Shugan (1997), “Film Critics: Influencers or Predictors?” Journal of Marketing, 61 (April), 68-78.
Epp, Amber M. and Linda L. Price (2008), “Family Identity: A Framework of Identity Interplay in Consumption Practices,” Journal of Consumer Research, 35 (June), 50-70.
Epstein, Edward J. (2010), The Hollywood Economist. Brooklyn, NY: Melville House Publishing.
Fromme, Johannes (2003), “Computer Games as a Part of Children’s Culture,” International Journal of Computer Game Research, 3 (1), 1-23.
Godes, David and Dina Mayzlin (2004), “Using Online Conversations to Study Word of Mouth Communications,” Marketing Science, 23 (4), 545-60.
Goldstein, Patrick and James Rainey (2009),“Is the 'Twitter Effect' on Box Office just Big Media Hype?”, (accessed August 17, 2010), [available at http://latimesblogs.latimes.com/the_big_picture/2009/10/is-the-twitter-effect-a-big-media-hype.html].
Granger, Clive W.J. (1969),"Investigating Causal Relations by Econometric Models and Cross-Spectral Methods,” Econometrica, 37 (3), 424–438.
Hair, Joseph F., William C. Black, Barry J. Babin, Rolph E. Anderson and Ronald L. Tatham (2006), Multivariate Data Analysis, Englewood Cliffs, NJ: Pearson Prentice Hall.
Hall, Mark, Eibe Frank, Geoffrey Holmes, Bernhard Pfahringer, Peter Reutemann, Ian H. Witten (2009), “The WEKA Data Mining Software: An Update” SIGKDD Explorations, 11 (1).
Hayes, Dade (2002), “Tentpoles Fade Fast After Opening Ad Blast,” Variety Magazine, August 4.
Hennig-Thurau, Thorsten, Kevin P. Gwinner, Gianfranco Walsh, and Dwayne D. Gremler (2004), “Electronic Word-of-Mouth via Consumer Opinion Platforms: What Motivates Consumer to Articulate Themselves on the Internet?” Journal of Interactive Marketing, 18 (1), 38-52.
36
———, Mark B. Houston, and Torsten Heitjans (2009), “Conceptualizing and Measuring the Monetary Value of Brand Extensions: The Case of Motion Pictures,” Journal of Marketing, 73 (November), 167-183.
Ho, Jason Y.C., Tirtha Dhar, and Charles B. Weinberg (2009), “Playoff Payoff: Super Bowl Advertising for Movies,” International Journal of Research in Marketing, 26, 168-179.
Holbrook, Morris B. (2005), “The Role of Ordinary Evaluations in the Market for Popular Culture: Do Consumers Have ‘Good Taste’?” Marketing Letters, 16 (2), 75–86.
——— and Michaela Addis (2007), “Taste versus the Market: An Extension of Research on the Consumption of Popular Culture,” Journal of Consumer Research, 34, 415–424.
Jansen, Bernard J., Mimi Zhang, Kate Sobel, and Abdur Chowdury (2009), “Twitter Power: Tweets as Electronic Word of Mouth,” Journal of the American Society for Information Science and Technology, 60 (11), 2169-88.
Jenkins, Roger L. (1979), "The Influence of Children in Family Decision-Making: Parents’ Perceptions," in Advances in Consumer Research, Vol. 6, William L. Wilkie, ed. Ann Arbor, MI: Association for Consumer Research, 413-18.
Kaplan, Andreas M. and Michael Haenlein (2011), “The Early Bird Catches the News: Nine Things You Should Know about Micro-Blogging,” Business Horizons, 54, 105-113.
Karniouchina, Ekaterina V. (2011), “Impact of Star and Movie Buzz on Motion Picture Distribution and Box Office Revenue,” International Journal of Research in Marketing, 28, 62-74.
Katz, Elihu and Paul F. Lazarsfeld (1955), Personal Influence: The Part Played by People in the Flow of Mass Communications. New York: The Free Press.
Keerthi, S. S., S. K. Shevade, C. Bhattacharyya, and K.R.K. Murthy (2001), “Improvements to Platt's SMO Algorithm for SVM Classifier Design,” Neural Computation, 13(3), 637-649.
Kirmani, Amna and Akshay R. Rao (2000), "No Pain, No Gain: A Critical Review of the Literature on Signaling Unobservable Product Quality," Journal of Marketing, 64 (2), 66-79.
Knowlton, Thomas (2011), “Twitter Grows, LinkedIn Stagnates, and Facebook Still Wears the Crown: A Social Networking Infographic,” (accessed November 16, 2011), [available at http://www.techvibes.com/blog/twitter-grows-linkedin-stagnates-and-facebook-still-wears-the-crown-a-social-networking-infographic-2011-09-07].
Lance, Charles E. (1988), “Residual Centering, Exploratory and Confirmatory Moderator Analysis, and Decomposition of Effects in Path Models Containing Interactions,” Applied Psychological Measurement, 12(2), 163-175.
37
Lang, Brent (2010), “Study: The 'Twitter Effect' Does Not Exist”, (accessed October 12, 2011), [available at http://www.thewrap.com/media/column-post/thegrill-twitter-effect-myth-21035?page=0,0].
Levine, Frederick, Christopher Locke, David Searls, and David Weinberger (2000), The Cluetrain Manifesto, Cambridge, MA: Basic Books.
Liu, Yong (2006), “Word-of-Mouth for Movies: Its Dynamics and Impact on Box Office Revenue,” Journal of Marketing, 70 (July), 74-89.
Mangleburg, Tamara F. (1990), "Children’s Influence in Purchase Decisions: A Review and Critique," in Advances in Consumer Research, Vol. 17, Marvin E. Goldberg, Gerald Gorn, and Richard W. Pollay, eds. Provo, UT: Association for Consumer Research, 813-25.
———, Patricia M. Doney, and Terry Bristol (2004), “Shopping with Friends and Teens’ Susceptibility to Peer Influence,” Journal of Retailing, 80 (2), 101-16.
Max Planck Institute (2011), “The Twitter Project Page at MPI-SWS,” Max Planck Institute for Software Systems, (accessed December 23, 2011), [available at http://twitter.mpi-sws.org/].
Mithas, Sunil and Mayuram S. Krishnan(2009), “From Association to Causation via a Potential Outcomes Approach,” Information Systems Research, 20(2), 295-313.
Moviemarketingmadness (2009), “Counteracting a Movie’s Negative Word-of-Mouth,” (accessed November 16, 2011), [available at http://www.moviemarketingmadness.com/blog/2009/04/counteracting-a-movie%E2%80%99s-negative-word-of-mouth/].
Mullaney, Timothy J. (2006), “Netflix - The Mail-Order Movie House that Clobbered Blockbuster,” (accessed October 12, 2011), [available at http://www.businessweek.com/smallbiz/content/may2006/sb20060525_268860.htm].
Pang, Bo, Lillian Lee, and Shivakumar Vaithyanathan (2002), “Thumbs Up? Sentiment Classification Using Machine Learning Techniques,” Proceedings of EMNLP, (accessed August 10, 2010), [available at http://www.cs.cornell.edu/home/llee/papers/sentiment.pdf].
Parr, Ben (2011a), “40% of All Tweets Come From Mobile,” (accessed November 11, 2011), [available at http://mashable.com/2011/01/07/40-of-all-tweets-come-from-mobile].
——— (2011b), “Twitter Has 100 Million Monthly Active Users; 50% Log In Every Day,” (accessed November 11, 2011), [available at http://mashable.com/2011/10/17/twitter-costolo-stats].
PBS (2001), “The Monster that Ate Hollywood,” PBS Frontline, (accessed December 28, 2011), [available at http://www.pbs.org/wgbh/pages/frontline/shows/hollywood/].
38
Platt, John C. (1999), “Fast Training of Support Vector Machines using Sequential Minimal Optimization,” in Advances in Kernel Methods - Support Vector Learning, Bernhard Schölkopf, Christopher Burghes, Alexander Smola, eds. Cambridge MA: MIT Press, 41-65.
Richins, Marsha L. (1983), “Negative Word-of-Mouth by Dissatisfied Consumers: A Pilot Study,” Journal of Marketing, 47 (Winter), 68-78.
Rosenbaum, Paul R. and Donald B. Rubin (1983), “The Central Role of the Propensity Score in Observational Studies for Causal Effects,” Biometrika, 70 (1), 41-55.
Singh, Jagdip (1990), “A Typology of Consumer Dissatisfaction Response Styles,” Journal of Retailing, 66 (1), 57-99.
Singh, Timon (2009), “How Hollywood Embraced Social Media,” (accessed December 28, 2011), [available at http://www.ngonlinenews.com/article/hollywood-and-social-media].
Van Grove, Jennifer (2009), “Did Opening Night Twitter Reviews Sink Bruno’s Weekend Box Office?,” (accessed November 16, 2011), [available at http://mashable.com/2009/07/13/bruno-twitter-reactions/].
Von Wangenheim, Florian and Tomàs Bayón (2007), “Behavioral Consequences of Overbooking Service Capacity,” Journal of Marketing, 71 (4), 36-47.
Zellner Arnold (1962), “An Efficient Method of Estimating Seemingly Unrelated Regression Equations and Tests of Aggregation Bias,” Journal of the American Statistical Association, 57, 500-9.
39
FIGURE 1 Types of Word of Mouth and Their Characteristics
TRADITIONAL WORD OF MOUTH (TWOM)
• Receiver is an individual person or a small group
• Real-time transmission• Personal influence
ELECTRONICWORD OF MOUTH (EWOM)
• Receiver is a potentially large group
• Asynchronous transmission• Anonymous
MICROBLOGGINGWORD OF MOUTH (MWOM)
40
FIGURE 2 Conceptual Model
Subsequent early product
adoption
MWOM valence(immediately after
release)H1
Customer segment susceptibility to
MWOM
H3
MWOM volume (immediately after
release)
H2
41
FIGURE 3 Distribution of Tweets throughout the Opening Weekend
Notes: All time data refer to Eastern Daylight Time.
0
10000
20000
30000
40000
50000
60000
70000Number of tweets per hour
Day/Time
Anticipatory, pre-consumption tweets
Evaluative, post-consumption tweets
42
TABLE 1 Variable Operationalizations
Variable Label Operationalization Data Source
Revenues on release day REV_REL North American box office revenues generated on Friday, in USD Boxofficemojo.com
Revenues on subsequent days of the opening weekend
REV_SUB Sum of North American box office revenues generated on Saturday and Sunday, in USD Boxofficemojo.com
MWOM valence MWOMVAL Quotient of positive/negative evaluative, post-purchase MWOM messages for a movie sent through Twitter between Friday, 2:00 p.m., and Saturday, 12:00 p.m. Eastern Daylight Time, classification based on sentiment analysis
Twitter.com API
MWOM volume MWOMVOL Number of evaluative, post-purchase MWOM messages for a movie sent through Twitter between Friday, 2:00 p.m., and Saturday, 12:00 p.m. Eastern Daylight Time
Twitter.com API
Family-targeted movie AUDFAM Main audience of movie is family (= 1, 0 otherwise) Jinni.com, own coding
Teenager-targeted movie AUDTEENS Main audience of movie is teenagers (= 1, 0 otherwise) Jinni.com, own coding
Pre-release movie buzz BUZZ Inverted rank in the Movie-Meter on IMDb at a movie’s release IMDb.com
Sequel SEQUEL Movie is a sequel (= 1, 0 otherwise) IMDb.com
Critics rating CRITRAT Average rating of a movie by up to 40 professional critics, weighted according to the influence of the experts as expressed by the Metascore (scale ranges from 1 to 10)
Metacritic.com
Starpower STAR Movie contains a major star (= 1, 0 otherwise) Quigley Publishing
Pre-release advertising spending
AD Advertising spending for a movie before its release, in USD Kantar Media
Production budget BUDGET Production budget of a movie, in USD (inflation corrected) IMDb/ Boxofficemojo
Ordinary evaluation ORDEVAL Average quality rating of a movie by users of Netflix (scale ranges from 1 to 5) Netflix.com
43
TABLE 2 Correlations and Descriptive Statistics
Mean SD 1 2 3 4 5 6 7 8 9 10 11
1 REV_SUB 16.171 .852 1.000
2 REV_REL 15.575 .899 .973 1.000
3 MWOMVAL 2.229 .571 .173 .139 1.000
4 MWOMVOL 7.587 1.446 .710 .787 .211 1.000
5 AUDFAM .180 .387 .148 .023 .105 -.204 1.000
6 AUDTEENS .530 .501 .082 .173 -.050 .234 -.502 1.000
7 BUDGET 3.696 .897 .676 .582 .076 .326 .236 -.010 1.000
8 CRITRAT 1.539 .319 .355 .342 .370 .369 .030 -.177 .258 1.000
9 BUZZ .529 .145 .671 .711 .061 .709 -.152 .194 .511 .291 1.000
10 STAR .290 .454 .224 .184 -.061 .031 -.133 -.127 .319 .164 .125 1.000
11 AD 9.770 .668 .442 .377 .001 .197 .103 -.044 .584 .184 .305 .309 1.000
12 SEQUEL .110 .320 .256 .310 .013 .173 .142 .036 .138 .040 .198 -.095 -.198 Notes: All variables (except dummies) are log-transformed, as used in the estimation process.
44
TABLE 3 Estimation Results (Seemingly Unrelated Regression)
Coef. Std. Err. z
DV = REV_SUB
REV_REL .944 .0274 34.41**
MWOMVAL .080 .0315 2.55*
MWOMVOL -.036 .0156 -2.34*
AUDFAM .292 .059 4.98**
AUDTEENS -.040 .050 -.80
BUDGET .115 .021 5.58**
CRITRAT .004 .044 .10
BUZZ -.116 .133 -.87
STAR .024 .030 .79
AD -.011 .023 -.46
SEQUEL -.183 .042 -4.41**
MWOMVAL×MWOMVOL .018 .011 1.62
MWOMVAL×AUDFAM -.009 .003 -2.62**
MWOMVAL×AUDTEENS -.000 .004 -.05
Constant 1.098 .381 2.89**
RMSE = .112; R2 = .98; Chi2 = 5763.12
DV = REV_REL
BUDGET .186 .087 2.14*
CRITRAT .317 .1798 1.76
BUZZ 3.113 .451 6.91**
SEQUEL .588 .186 3.17**
STAR .046 .129 .36
AD .176 .107 1.64
Constant 10.957 .923 11.87**
RMSE = .552; R2 = .62; Chi2 = 172.21
Notes: RMSE = root mean standardized error; ** p < .01, * p < .05
45
TABLE 4
Probit Regression Results DV = MWOM period dummy Coef. Std. Err. z
BUDGET .001 .002 .33
AD .000 .000 1.23
BUZZ -1.915 .533 -3.59**
CRTIRAT -.1207 .043 -2.83**
ORDEVAL 1.014 .107 9.47**
SEQUEL -.060 .195 -.31
STAR .043 .143 .30
AUDFAM .091 .184 .49
AUDTEENS .564 .136 4.14**
Constant -8.259 .700 -11.80**
Likelihood regression Chi2(9) = 155.11; Prob > Chi2 < .00; Pseudo R2 = .22; Log-likelihood = –278.685.
Notes: ** p < .01, * p < .05
46
TABLE 5 Comparison of Sample Differences Before and After Matching
Variable Mean t-Test
Sample Treated Control %bias %reduct |bias| t BUDGET Unmatched 58.903 50.896 16.9 1.97* Matched 58.903 60.985 -4.4 74.0 -.31 AD Unmatched 20,230 17,506 33.9 3.52** Matched 20,230 21,519 -16.0 52.7 -1.17 BUZZ Unmatched .067 .127 -32.5 -2.84** Matched .067 .099 -17.5 46.1 -1.38 CRITRAT Unmatched 4.895 4.680 13.0 1.20 Matched 4.895 5.125 -13.9 -7.0 -1.06 ORDEVAL Unmatched 7.421 6.698 95.7 9.52** Matched 7.421 7.360 8.1 91.6 .65 SEQUEL Unmatched .114 .116 -.5 -.05 Matched .114 .048 20.8 -439.6 1.78 STAR Unmatched .286 .282 .9 .09 Matched .286 .2952 -2.1 -135.9 -.15 AUDFAM Unmatched .181 .138 11.8 1.22 Matched .181 .133 13.0 -1.0 .95 AUDTEENS Unmatched .533 .369 33.4 3.32** Matched .533 .562 -5.8 82.6 -.41
Notes: ** p < .01, * p < .05
47
TABLE 6 Seemingly Unrelated Regression Results for MWOM Period Sample and Matched Pre-MWOM Period Samples
MWOM Period Sample Pre-MWOM Period Sample Coef. Std. Err. z Coef. Std. Err. z P>|z| DV = REV_SUB REV_REL .868 .020 43.38** .968 .034 28.4** .000 AUDFAM .200 .042 4.80** .203 .059 3.43** .001 AUDTEENS .036 .031 -1.16 -.056 .045 -1.24 .214 BUDGET .112 .021 5.39** .0485 .039 1.26 .209 CRITRAT .022 .044 -.50 .090 .059 1.52 .127 BUZZ .118 .184 -.64 .307 .1912 1.61 .107 SEQUEL .157 .044 -3.59** -.058 .083 -.69 .488 STAR .048 .030 1.57 .146 .044 3.31** .001 AD .002 .024 .07 -.165 .072 -2.28* .022 ORDEVAL .326 .129 2.52* -.002 .281 -.01 .994 Constant 1.646 .364 4.53** 2.482 .701 3.54** .000 RMSE = .122; R2 = .98; Chi2= 4943.98 RMSE = .176; R2 = .95; Chi2 = 2198.99 DV = REV_REL
BUDGET .283 .096 2.95** .236 .109 2.17* .030 CRITRAT .444 .200 2.22* .177 .157 1.13 .259 BUZZ 3.447 .854 4.04** 1.175 .569 2.06* .039 SEQUEL .634 .209 3.03** .489 .255 1.92 .055 STAR .024 .145 .16 -.164 .135 -1.21 .225 AD .209 .120 1.74 1.070 .199 5.38** .000 Constant 10.181 1.057 9.63** 3.073 1.764 1.74 .081 RMSE = .620; R2 = .52; Chi2 = 113.87 RMSE = .548; R2 = .53; Chi2 = 120.57 Notes: RMSE = root mean standardized error; ** p < .01, * p < .05
48
APPENDIX
Movie Titles (MWOM Sample)
TITLE
RELEASE DATE
2012 13-Nov-09A Christmas Carol 6-Nov-09Alice In Wonderland 5-Mar-10Alpha and Omega 17-Sep-10Amelia 23-Oct-09Armored 4-Dec-09Astro Boy 23-Oct-09Avatar 18-Dec-09Brooklyn's Finest 5-Mar-10Case 39 1-Oct-10Cats & Dogs: The Revenge of Kitty Galore 30-Jul-10Charlie St. Cloud 30-Jul-10Cirque du Freak: The Vampire's Assistant 23-Jan-09Clash of the Titans 2-Apr-10Cop Out 26-Feb-10Couples Retreat 9-Oct-09Date Night 9-Apr-10Daybreakers 8-Jan-10Dear John 5-Feb-10Despicable Me 9-Jul-10Devil 17-Sep-10Diary Of A Wimpy Kid 19-Mar-10Did You Hear About the Morgans? 18-Dec-09Dinner for Schmucks 30-Jul-10Easy A 17-Sep-10Edge of Darkness 29-Jan-10Everybody's Fine 4-Dec-09Extraordinary Measures 22-Jan-10From Paris with Love 5-Feb-10Furry Vengeance 30-Apr-10Get Him to the Greek 4-Jun-10Going the Distance 3-Sep-10Green Zone 12-Mar-10Grown Ups 25-Jun-10Hot Tub Time Machine 26-Mar-10How To Train Your Dragon 26-Mar-10Inception 16-Jul-10Invictus 11-Dec-09Iron Man 2 7-May-10It's Complicated 25-Dec-09Just Wright 14-May-10Leap Year 8-Jan-10Legend of the Guardians: The Owls of Ga'Hoole 24-Sep-10Legion 22-Jan-10Let Me In 1-Oct-10Letters To God 9-Apr-10Letters to Juliet 14-May-10Life As We Know It 8-Oct-10Lottery Ticket 20-Aug-10MacGruber 21-May-10Machete 3-Sep-10Marmaduke 4-Jun-10
TITLE
RELEASE DATE
Nanny McPhee Returns 20-Aug-10New Moon 20-Nov-09Nightmare On Elm Street 30-Apr-10Our Family Wedding 12-Mar-10Percy Jackson & the Olympians: The Lightning Thief 12-Feb-10Piranha 3-D 20-Aug-10Pirate Radio 13-Nov-09Planet 51 20-Nov-09Predators 9-Jul-10Prince of Persia: The Sands of Time 28-May-10Ramona and Beezus 23-Jul-10Remember Me 12-Mar-10Repo Men 19-Mar-10Resident Evil: Afterlife 10-Sep-10Robin Hood 14-May-10Salt 23-Jul-10Saw VI 23-Oct-09Secretariat 8-Oct-10Sherlock Holmes 25-Dec-09She's Out Of My League 12-Mar-10Shrek Forever After 21-May-10Shutter Island 19-Feb-10Step Up 3-D 6-Aug-10Takers 27-Aug-10The A-Team 11-Jun-10The Back-up Plan 23-Apr-10The Blind Side 20-Nov-09The Book of Eli 15-Jan-10The Bounty Hunter 19-Mar-10The Box 6-Nov-09The Crazies 26-Feb-10The Fourth Kind 6-Nov-09The Karate Kid 11-Jun-10The Last Exorcism 27-Aug-10The Losers 23-Apr-10The Lovely Bones 15-Jan-10The Men Who Stare at Goats 6-Nov-09The Other Guys 6-Aug-10The Social Network 1-Oct-10The Spy Next Door 15-Jan-10The Stepfather 16-Oct-09The Switch 20-Aug-10The Town 17-Sep-10The Wolfman 12-Feb-10Tooth Fairy 22-Jan-10Toy Story 3 18-Jun-10Wall Street: Money Never Sleeps 24-Sep-10When in Rome 29-Jan-10Where the Wild Things Are 16-Oct-09Why Did I Get Married Too? 2-Apr-10You Again 24-Sep-10Youth in Revolt
8-Jan-10