This article appeared in a journal published by Elsevier. The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright
Author's personal copy
Intern. J. of Research in Marketing 26 (2009) 168–179
Contents lists available at ScienceDirect
Intern. J. of Research in Marketing
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / i j r e s m a r
Playoff payoff: Super Bowl advertising for movies☆
Jason Y.C. Ho a,⁎, Tirtha Dhar b, Charles B. Weinberg b
a
b
Simon Fraser University, 8888 University Drive, Burnaby, B.C., Canada V5A 1S6
Sauder School of Business, University of British Columbia, 2053 Main Mall, Vancouver, B.C., Canada V6T 1Z2
a r t i c l e
i n f o
Article history:
First received in 28 July 2008
and was under review for 6 months
Area Editor: Richard Staelin
Keywords:
Super Bowl
Advertising
Movies
Marketing channel
a b s t r a c t
Marketers are increasingly making use of major TV events, such as the Super Bowl, to advertise their
products. However, the economic value of such advertising is highly uncertain. Since an ad during the Super
Bowl costs 2.5 times more per viewer reached than an ad during a network TV prime time show, developing
methods for evaluating such advertising and for measuring its effects seems particularly important. Using the
setting of the movie industry, this paper develops and estimates a model that includes both direct (on
potential moviegoers) and indirect effects (on exhibitors) of regular and Super Bowl advertising. The model
recognizes the endogeneity of advertising, and in particular develops a discrete choice model to control for
the endogeneity of the Super Bowl advertising decision. The results indicate that Super Bowl advertising has
a positive effect on box office revenues, but primarily through an indirect effect on exhibitors. In addition,
regular TV advertising is more effective than Super Bowl advertising for initial advertising spending; a
counterfactual analysis, by contrast, shows that for a movie already spending at our sample's average TV
spending level of $13 million, Super Bowl advertising has a greater effect on revenues than regular TV
advertising.
© 2009 Elsevier B.V. All rights reserved.
1. Introduction
Marketers, confronted on one hand by the increasing ability of consumers to avoid watching commercials aired during TV shows and on
the other by the sheer clutter of TV advertising, have turned both to
“stealth marketing,” where their presence is largely hidden, and to major
TV event advertising, where their presence is written bold. While an
increasing number of papers look at the effects of stealth marketing (e.g.,
Russell, 2002; Mayzlin, 2006), little is known about major event
advertising. Of all of the major TV events in the U.S., the annual broadcast
of the Super Bowl (SB) is the most anticipated, discussed, and expensive.
In this paper, our focus is on evaluating advertising during the Super
Bowl, an iconic American event with more than 90 million viewers each
year — the most viewed TV show in the US. The second most viewed
show, the Academy Awards, attracts about half that number of viewers.
To place a 30-second ad during the Super Bowl, marketers have to pay
more than $2.3 million. Despite such high cost, neither industry
(Advertising Age, Jan 31, 2005) nor academia can provide much insight
on the value of advertising during the Super Bowl in particular and major
TV events in general.
☆ The authors gratefully acknowledge the support of the Social Sciences and
Humanities Research Council of Canada.
⁎ Corresponding author.
E-mail addresses: [email protected] (J.Y.C. Ho), [email protected] (T. Dhar),
[email protected] (C.B. Weinberg).
0167-8116/$ – see front matter © 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijresmar.2009.06.001
A Super Bowl advertisement is not only the most expensive spot on
TV in absolute terms, but is also 2.5 times more expensive per viewer
reached than regular advertising. In 2004, a 30-second prime time
network TV commercial cost approximately $120,500, with a cost per
thousand viewers (CPM) of $19.85, compared to the estimated
$2.3 million cost of an advertisement during the 2004 Super Bowl,
with a CPM of $51.26. Given these cost economics, are there circumstances under which a Super Bowl advertisement is a better investment
than an advertisement aired during a regular TV show?
To answer this question, we examine the market impact of Super
Bowl advertising for movies. Using data on the U.S. movie industry from
2000 to 2002, we build a system of equations model to study the
potential effects of Super Bowl advertising on both movie exhibitors and
moviegoers. Our empirical results demonstrate that:
1. Super Bowl advertising for a movie influences the opening week
box office revenues by indirectly attracting more movie exhibitors
to show the movie, thus increasing product availability, which in
turn increases initial box office revenues. Super Bowl advertising
does not directly affect the moviegoers in the opening week (or in
subsequent weeks).
2. Super Bowl advertising is not as effective as other TV advertising
expenditures prior to the movie's release if both types of prelaunch TV advertising expenditures are evaluated at the same
initial levels. On the other hand, given the presence of the welldocumented diminishing returns to scale effect for advertising (see
Vakratsas & Ambler (1999) for a review), we found that for a movie
Author's personal copy
J.Y.C. Ho et al. / Intern. J. of Research in Marketing 26 (2009) 168–179
169
Fig. 1. Pre-launch advertising patterns of The Mummy Returns and U-571.
that has already allocated the average amount of approximately
$13 million that a mass market movie in our sample spends on TV
advertising, spending about $2.2 million on Super Bowl advertising
is usually more effective than adding the same amount of money to
spending on regular TV advertising.
2. Related literature
Our goal in this paper is to use Super Bowl advertising by the movie
industry as a context in which to explore the unique role of major event
advertising in a marketing channel after taking into account some of the
unintentional shortcomings of earlier studies.
2.1. Major TV event advertising
Despite being an important marketing tactic, the value of major TV
event advertising has not received much academic attention. With
regard to the Super Bowl, the only major TV event that has been studied,
most studies (such as Pavelchak, Antil, & Munch, 1988; Newell &
Henderson, 1998; Newell, Henderson, & Wu, 2001; Tomkovick, Yelkur, &
Christians, 2001) have primarily used the Super Bowl as a field setting to
examine the effects of various design and media factors in advertising.
An exploratory study by Yelkur, Tomkovick, and Traczyk (2004), also
focusing on the movie industry, appears to be the sole exception.
Compared to that study, we have developed a richer data set, an improved methodology and a more sophisticated model to describe how
Super Bowl advertising works through downstream channel members
to impact final consumer demand.
The lack of research on the value of major TV event advertising is
partly due to data challenges: in most major product categories, only
one or two brands advertise during a specific major TV event. In
addition, it is difficult to distinguish continuing sales from incremental
sales due to major event advertising. In light of these data challenges
and research issues, we focus on the U.S. movie industry. In the three
year period (2000–2002) that we examine, there were 19 different
movies advertised during the three Super Bowl games.1 Only beer and
soft drink companies' brands had more Super Bowl advertisements
than did movies. Unlike these products, all of the advertisements in the
context of our study were for yet-to-be released movies, thereby helping us to avoid the seemingly intractable challenge of disentangling
the effect of regular advertising on continuing sales from incremental
sales due to major event advertising.
2.2. Mediating role of downstream channel members in advertising effects
Although marketers primarily use advertising to stimulate consumers' demand (i.e., the “pull” effect), some advertising campaigns can
also have a “push” effect on retailers (Montgomery, 1975; Olver & Farris,
1989; Chu, 1992; Desai, 2000), and the increased product availability at
the retail level can then increase consumer adoption (Jones & Ritz,1991).
Such “pull and push” effects have been noted by practitioners.2 However,
only a few empirical studies have formally examined such dual effects:
1
The percentage of observations possessing our focal characteristic, Super Bowl
advertising, is similar to the percentage in previous studies in a similar context. For
example, Basuroy et al. (2006) have about 6% of their sample being sequels, their focal
variable.
2
When Master Lock started its three-decade-long series of Super Bowl advertisements in 1974, the primary target was not consumers, but distributors (Kanner, 2004).
More recently, to explain how its commercial during Super Bowl 2005 for Emerald of
California nuts resulted in a 56% sales increase in the four weeks after the Super Bowl,
the firm's marketing director said, “We are a new brand in a very, very tough category,
and being on the Super Bowl was a great way to tell consumers and retailers that we
are here to stay.” (New York Times, Jan 18, 2006).
Author's personal copy
170
J.Y.C. Ho et al. / Intern. J. of Research in Marketing 26 (2009) 168–179
Fig. 2. Dual path model of Super Bowl TV advertising.
Elberse & Eliashberg (2003; hereafter EE) and Basuroy, Desai, & Talukdar
(2006); hereafter BDT) support the view that advertising positively
affects distribution intensity, which in turn increases final consumer
sales. Extending these two studies, which only consider aggregate
advertising spending, we decompose advertising spending into Super
Bowl and regular advertising and compare the effectiveness of these two
different advertising tactics.
Fig. 1 illustrates the advertising patterns prior to the release weeks
for two movies, U-571 and The Mummy Returns. As can be seen, Super
Bowl advertising tends to be separate from the overall pre-launch
advertising campaign; U-571's Super Bowl advertising appeared
12 weeks before its release, and The Mummy Returns', 14 weeks
before. The average start time for the Super Bowl advertised movies in
our sample was 13 weeks before launch, as compared to 4 weeks for
the start of national advertising for non-Super Bowl advertised movies
in our sample. Thus, Super Bowl advertising not only involves large
sums of money, but is clearly distinct from regular advertising.
2.3. Endogeneity of Super Bowl advertising
The simultaneity (and in some cases reverse causality) of advertising
and sales has been well documented in the advertising literature (e.g.,
Ashley, Granger, & Schmalensee, 1980; Heyse & Wei, 1985). To measure
the true effect of advertising on sales, we thus need to control for the
endogeneity of advertising. Extending EE (2003), which only models the
endogeneity of distribution channels, BDT (2006) captures the endogeneity of both distribution channels and advertising. We further
extend their studies by capturing the endogeneity of the Super Bowl
advertising decision process. This is an important methodological extension because the Super Bowl advertising spending variable, from the
perspective of a statistician, is similar to a binary choice variable; all
Super Bowl advertised movies have a spending level of about $2 million,
while the non-Super Bowl movies have zero values. We test and correct
for the bias of the endogeneity of Super Bowl advertising by estimating
its effect with a two-stage instrumental variable estimator modified for a
discrete endogenous decision variable following Mroz (1999). Our
approach is in contrast with other studies also controlling for a discrete
endogenous variable, which usually use a linear regression model to
approximate the binary endogenous variable (e.g., Leenheer, Van
Heerde, Bijmolt, & Smidts, 2007). In addition, as detailed below,
extending EE (2003) and BDT (2006), we include advertising lead
time and direct measures of pre-launch word-of-mouth and publicity to
better control for the effects of these other communication variables on
sales, thus further clarifying Super Bowl advertising's role in the
marketing process.
3. Model development
To capture the potential “pull” and “push” effects of Super Bowl
advertising for movies, we propose a dual path model. Fig. 2 depicts
the main model characterizing the relations among Super Bowl
advertising, regular launch TV advertising, distribution coverage, and
consumer purchases. Consistent with EE (2003) and BDT (2006), we
hypothesize that regular launch TV advertising has the same dual
paths as Super Bowl advertising does.3 For brevity, we hereafter
discuss only the dual paths of Super Bowl advertising. There are two
paths by which Super Bowl advertising can influence consumer demand, namely a direct advertising effect on consumer demand and an
indirect effect through downstream channel members.
We formalize the key features of our dual path model in the following two hypotheses:
H1. Super Bowl advertising increases the opening week box office
revenues along two paths:
H1a (Direct Path). Super Bowl advertising directly increases
opening week box office revenues after controlling for the mediation of the movie exhibitors.
H1b (Indirect Path). Super Bowl advertising first increases the
number of movie exhibitors. Then, the increased number of movie
exhibitors increases opening week box office revenues.
H2. Super Bowl advertising is as effective as regular TV advertising.
Although we model only first week sales to focus on the initial effects
of Super Bowl advertising, we also analyzed the effects of Super Bowl
advertising on subsequent week sales. Qualitatively, the results for Super
Bowl advertising do not change from those for the first week, as discussed below.4
3.1. Econometric model
To test the key feature in our dual path model, we develop the
following system of equations model relating opening week box office
revenues for movie j (denoted as BOj), opening week number of
theaters engaged for movie j (denoted as THEATERj), regular launch
3
Essentially, we use TV advertising expenditure as a proxy for total advertising
expenditure in the pre-launch period. According to 2003 MPAA market statistics, TV
advertising was the major medium used by movie distributors from 2000 to 2002.
4
Detailed results for this model are available from the authors upon request.
Author's personal copy
J.Y.C. Ho et al. / Intern. J. of Research in Marketing 26 (2009) 168–179
TV advertising expenditure by movie j (denoted as TVADj), and Super
Bowl advertising expenditure by movie j (denoted as SUPERBOWLj):
β1:0
THEATERj = e
β
β
⋅SUPERBOWLj 1;SB ⋅TVADj 1;AD
ð1Þ
∑β1;h ⋅ZThj
β
ε
⋅∏XTkj1;k ⋅e ∀h
⋅e 1
∀k
β 2:0
BOj = e
⋅∏X Bkj
∀k
β
β
β
⋅SUPERBOWLj 2;SB ⋅TVADj 2;AD ⋅THEATERj 2;THR
β 2;k
ð2Þ
∑β2;h ⋅ZBhj
ε
⋅e ∀h
⋅e 2
where X and Z are movie characteristics to be defined below, ε1 and ε2
are the errors of the two equations, and the βs are the parameters to be
estimated.
Unlike such previous studies as EE (2003) and BDT (2006), we
decompose total TV advertising spending into Super Bowl (SUPERBOWLj)
and regular launch TV advertising expenditure (TVADj). Parameters β2,SB
and β2,AD in the box office Eq. (2) capture the direct effects of Super Bowl
and regular advertising, respectively, while parameters β1,SB and β1,AD
in Eq. (1) and β2,THR in Eq. (2) together characterize the indirect paths.
Such decomposition allows us to examine the relative effectiveness of
Super Bowl and regular TV advertising spending by comparing the
parameter estimates β2,SB versus β2,AD and β1,SB versus β1,AD.
To capture the well-documented diminishing marginal returns effect
in advertising (Vakratsas & Ambler, 1999), both Eqs. (1) and (2) are
multiplicative, which allows either a concave (diminishing return to
scale) or convex shape (increasing return to scale) for continuous
variables. In other words, comparing β2,SB vs. β2,AD and β1,SB vs. β1,AD
allows us to evaluate the differences between the marginal effects of
Super Bowl and regular TV advertising at the same initial level, say both
being zero, and in a more realistic scenario, where an average movie
spends approximately $13 million on regular advertising but zero on
Super Bowl advertising. Such a comparison, anticipating our empirical
results, includes the effect of diminishing returns to scale of regular TV
advertising.
While our model specification is similar in spirit to that of BDT
(2006), in addition to our decomposition of advertising expenditures,
it also departs from the usage of expected box office revenue as one of
the key explanatory variables in the theater equation. Specifically, BDT
(2006) used realized total revenue as a proxy for expected total
revenue in the first week theater equation. In this paper, we specify
the theater number as a function of a comprehensive set of available
characteristics in order to avoid some of the potential pitfalls of using
expected revenue as an explanatory variable.5
XTkj and ZThj are the characteristics potentially influencing theater
managers' screening decisions, with the former being a set of continuous
variables and the latter being a set of indicator (binary) variables. Similarly, XBkj and ZBhj are the continuous and indicator variables potentially
affecting moviegoers' ticket purchase decisions. A majority of these Xkj
and Zhj variables are common across Eqs. (1) and (2). These common
variables are competitive intensity for movie j (COMPj), buzz (BUZZj),
lead time of the launch advertising campaign (LEADj), amount of
publicity generated (PUBj), movie j being a sequel (SEQj), movie j being
of a certain genre (GENREj), movie j's rating by MPAA (MPAAj), star
5
We avoid using one single measure of expected total revenue for two reasons: [1] if
the proxy for the expected box office revenue is highly correlated with theaters (in
other words, theater owners are very good at predicting movie outcomes), then the
significance of the variables of interest will diminish substantially; and [2] similarly, if
other explanatory variables influence the expected revenue (in other words, theater
managers form their expectations based on available movie characteristics), then the
model will be overspecified, leading to a possible decrease in significance of the rest of
the explanatory variables in the theater equation.
171
Table 1
Definitions of variables.
Treated as endogenous variables
BOj
Total box office receipts from Friday to Thursday for movie j in
the release week
THEATERj
Number of movie theaters engaged for movie j in the release week
SUPERBOWLj Super Bowl TV advertising expenditure by movie j
TVAD,j
Total regular TV advertising expenditure up to and including the
release week of movie j
Treated as exogenous variables
COMPj
Total production budgets of all movies released in the same week
and one week prior to the release of movie j
BUZZ,j
Buzz for movie j in its release weeka
LEADj
Number of weeks between the first major TV ad and the release
week of movie j
PUB,j
Cumulative publicity received by movie j, up to and including
the release weekb
SEQj
A binary variable to indicate if movie j is a sequel
GENREj
Binary variables to indicate the genre: 1) action, 2) comedy, 3)
drama, and 4) familyc
MPAAj
Binary variables to indicate the MPAA rating: 1) G, 2) PG, 3) PG-13, and 4) R.
S_POWERj
A binary variable to indicate if either of the two major actors of Movie j
was on the previous year's Entertainment Weekly Power List Top 50.
D_POWERj
A binary variable to indicate if any of the directors of Movie j was on
the previous year's Entertainment Weekly Power List Top 50.
BUDGETj
Production budget of movie jd
RUNTIMEj
Runtime of movie j
DISTRIBUTORj Binary variables to indicate if movie j is distributed by one of the
following distributors: 1) Disney, 2) AOL, 3) Viacom, 4) Sony, 5)
20th Century Fox, 6) Vivendi, 7) DreamWorks, and 8) Other
movie distributors
CRITICSj
Average critics' rating given for movie je
SEASONj
A set of binary variables to indicate if movie j is released in one of the
following four Hollywood seasons: 1) January–April, 2) May–August,
3) September–October, and 4) November–December
HOLIDAYj
A binary variable to indicate if movie j is released in the week of a
major U.S. holiday
a
The buzz measure is an inverse transformation of MOVIEmeter from IMDb.com. Based
on which specific movie pages its four to five million weekly visitors view, IMDb.com
produces the weekly MOVIEmeter ranking for more than 290,000 movie titles in its
database. However, IMDb.com does not provide weekly traffic data at its site. So before the
inverse transformation, the weekly rankings are adjusted for each week's web traffic using
estimates from Alexa.com. The detailed derivation of the buzz variable is in Appendix A.
b
We measure the publicity for a movie by determining the total amount of coverage the
movie received in Entertainment Weekly, which has a circulation of 1.79 million, the largest
after the number one publication, TV Guide, in the entertainment magazine category (Audit
Bureau of Circulations). We first identified articles related to specific movies by coding the
table of contents of each issue of Entertainment Weekly. We then classified each article into
one of ten categories, e.g., Departments, News & Notes Category I, or Movie Review. We
determined the amount of publicity generated by each article using the average number of
pages of the category to which it belongs. PUBj is then defined as the sum of coverage values
of all articles for movie j before and including its release week.
c
Starting from the twelve genre categories used by Variety.com, we simplified the
categories into four main types, namely Action, Comedy, Drama and Family.
d
We obtained the estimated production budgets from IMDb.com.
e
Average critic ratings for individual movies were collected from RottenTomatoes.com.
power (S_POWERj), and director power (D_POWERj). Table 1 provides
detailed definitions and information on the data sources.
Some variables are specific to each equation. Here we provide
reasons for their inclusion in each equation. The following three variables are unique to the theater equation (i.e., Eq. (1)): [1] Total production budget for movie j (BUDGETj): As theater managers are much
better informed about movie budgets than typical consumers, we use
production budget in the theater Eq. (1), but not in the box office
Eq. (2).6 [2] Run time of movie j (RUNTIMEj): Ceteris Paribus, theater
owners prefer shorter movies over longer ones because this allows them
6
We also estimated a model with BUDGETj in both equations, but the effect of
BUDGETj on the box office revenues is not significant at the 5% level. More importantly,
the estimates of all of our focal variables are qualitatively the same as those in the
model with BUDGETj excluded from Eq. (2). Our treatment of budget is similar to that
of EE (2003), and they also did not find budget to be significant for the US first week
theater engagement equation.
Author's personal copy
172
J.Y.C. Ho et al. / Intern. J. of Research in Marketing 26 (2009) 168–179
to schedule more movies and reduces the operational costs of longer
screening times. On the other hand, anecdotal evidence suggests that
most consumers are not aware of the length of the movies, and, as a
result, this should not impact movie sales. [3] Distributor/studio of
movie j (DISTRIBUTORi): Different movie distributors may have differential strategies and power in dealing with movie theaters. As such
effects are largely confined to exhibitors, we include the distributor
variables only in the theater equation.
Three sets of variables are unique to Eq. (2), the box office equation:
[1] Critics' ratings of movie j (CRITICSj): As critics' ratings are not
available until just before a movie's release, theater managers cannot
use critics' ratings in making their screening decisions, which need to
be finalized at least a week prior to the movies' releases. [2] Movie
season in which movie j is released (SEASONj): Einav (2007) argues
that seasonality can only influence demand, but not the number of
available theaters for movie screening. This is due to the fact that the
number of theaters does not change from season to season. [3] Movie j
released during one of the eight major U.S. holidays (HOLIDAYj): similar
to the movie season variable, we expect HOLIDAYj to affect only the box
office revenues.
Table 2
Summary statistics of Non-Super Bowl (NSB) vs. Super Bowl (SB) advertised movies.
Variables
Group
Mean
Std.
Deviation
Minimum
Maximum
Box office revenues
($ in millions)
Number of theaters
(in thousands)
Regular TV advertising
spending ($ in millions)
Production budget
($ in millions)
Advertising lead time
(in weeks)
Movie runtime
(in Minutes)
Competition intensity
($ in millions)
Pre-launch buzz
SB Movies
NSB Movies
SB Movies
NSB Movies
SB Movies
NSB Movies
SB Movies
NSB Movies
SB Movies
NSB Movies
SB Movies
NSB Movies
SB Movies
NSB Movies
SB Movies
NSB Movies
SB Movies
NSB Movies
SB Movies
NSB Movies
42.81
23.89
2.97
2.49
14.44
13.07
70.00
50.90
15.84
5.48
114.00
107.29
154.10
190.94
4.45
4.06
4.74
2.87
5.25
5.14
29.32
23.09
0.37
0.59
3.38
4.44
22.10
29.65
9.89
4.63
16.65
18.35
51.79
89.40
1.01
1.01
5.58
4.06
0.94
1.31
10.54
1.80
2.23
0.75
8.53
0.21
17.00
15.60
3
0
94
72
63.40
9.50
0.37
0.01
0.00
0.00
3.90
2.20
110.56
151.62
3.61
3.68
20.53
23.98
103.00
142.00
39
34
155
183
285.00
456.00
5.12
5.26
21.70
31.55
7.00
8.50
Publicity
Critics' rating
3.2. Estimation challenges
To estimate Eqs. (1) and (2) simultaneously, we use log transformation to linearize the model. We have five endogenous variables on
the right hand side of the equations: TVADj, and SUPERBOWLj in the
theater equation and THEATERj, TVADj, and SUPERBOWLj in the box
office equation. The main sources of endogeneity are the potentially
unobserved (i.e., unobserved by the researchers) characteristics influencing the behaviors of the decision makers in a market: [1] a movie
distributor/studio with a movie of certain unobserved characteristics
is more likely to adopt Super Bowl advertising and/or commit to a
certain level of regular advertising; [2] theater managers are more
likely to screen a movie having these unobserved characteristics, and
[3] moviegoers are more likely to watch a movie having these unobserved characteristics in the opening week. An example of such
unobserved movie characteristics is the level of special effects in a
movie. As this characteristic is unobserved to us and thereby absent
from our model, there is a potential bias in the estimated effects of the
endogenous variables.7
We address the potential endogeneity of our focal variables in two
ways. First, we include a comprehensive set of observable movie
characteristics in our model (see the above discussion of X's and Z's).
These observable movie characteristics are usually the cues theater
managers and consumers use to infer a new movie's appeal. Second,
we control for any other unobserved effects of movie characteristics
on our focal variables by using a two-stage instrumental variable
estimation process. We discuss the details of our estimation procedure
in Section 5.
4. Data description
During the period of our study (i.e., 2000–2002) 1445 movies were
released. The majority of these movies received limited releases,
implying that they were shown in only a few local markets in the first
week. None of the Super Bowl movies during this period had a limited
release; all received a wide release across US. This is expected given
the magnitude of the Super Bowl expenditure and its broad reach.
Consequently, given that only movies with a major national release
are likely to advertise during the Super Bowl, we limit our analysis to
wide-release movies. Following Einav (2007), we chose movies
released in at least 600 theaters, leaving 402 movies in the sample.
7
For further discussion on omitted variables and endogeneity, please refer to Greene
(2003, page 334).
Even within these 402 movies, there is still great variation in the
number of theaters in which the movies were screened.
The second sampling criterion is the film's production budget. As
movies with low production budgets are unlikely to be able to invest
in Super Bowl advertising, we dropped them from our analysis. In
particular, our sample consists of movies with a production budget of
$15 million or more.8 We also dropped four movies due to missing
data on production budgets, resulting in a sample of 302 movies. For
these 302 movies, we still observe substantial variation in production
budgets. In sum, our sample accounts for 79% of the total North
American box office revenues ($25 billion) for all movies released
from 2000 to 2002.9
We identified the TV commercials for individual movies placed in
the Super Bowls of 2000, 2001 and 2002 from TV recordings of the
Super Bowl games from kickoff to the end of the game. While there
were TV commercials for other films appearing before and after the
games (e.g., during the pre-game show), we included only the
commercials appearing in the commercial breaks during the games.
This is the standard definition of Super Bowl advertising in the
academic and trade literature. There were 19 movies advertised
during the Super Bowls from 2000–2002.10 While all of the Super
Bowl-advertised movies placed a single 30-second commercial during
the Super Bowl (except Mission to Mars, which aired a 60-second
advertisement), there is clear variation in such other variables as
production budget and launch TV advertising spending.
As shown in Table 1, the variables in our study are constructed from
several different data sources. The major sources are: [1] Variety.com,
the website of the industry's authoritative trade magazine; [2] IMDb.
com, the popular interactive movie database website visited by more
than 25 million visitors each month; [3] TNS/CMR, the research
company tracking TV commercials on over 425 network and cable
channels in more than 75 TV markets in the United States; [4] Entertainment Weekly, the popular consumer magazine for entertainment;
[5] Rottentomatoes.com, a comprehensive website archiving reviews
by movie critics; and [6] Alexa, an Amazon.com subsidiary that tracks
web surfing of Internet users based on their proprietary search tool
8
In our sample of movies that advertised during the Super Bowl, the minimum
number of theaters in which any movie was released was 2225 and the minimum
production budget was $17 million, both for 40 Days and 40 Nights.
9
Our sampling criteria are different from those used by EE (2003) and BDT (2006):
EE selected only movies with top 10 box office results, while BDT selected movies
reviewed by selected critics in Variety.
10
Details of the Super Bowl movies are available from the authors upon request.
Author's personal copy
Table 3
Correlation matrix.
Variables
1
2
3
4
20
21
22
23
24
25
26
27
28
29
30
31
1
2
3
4
5
6
7
8
9
1.00
0.02
−0.01
0.08
−0.06
0.00
−0.11
0.06
−0.03
0.00
0.03
−0.10
0.05
−0.02
0.02
1.00
0.38
0.34
0.17
0.37
0.22
−0.29
−0.01
0.11
0.04
−0.10
−0.04
0.02
−0.01
1.00
−0.04
0.05
0.31
0.07
−0.31
0.43
−0.18
0.05
0.18
−0.12
0.10
−0.01
1.00
0.06
0.15
0.18
− 0.08
− 0.23
0.16
0.01
− 0.14
− 0.02
− 0.01
− 0.09
1.00
0.06 1.00
0.06 0.01 1.00
− 0.07 −0.10 −0.42 1.00
0.05 0.10 −0.40 −0.48 1.00
− 0.02 0.04 −0.12
0.04 −0.21 1.00
− 0.02 −0.04 0.14
0.12 −0.07 − 0.37 1.00
0.08 0.07 0.01 −0.15
0.28 − 0.26 − 0.70 1.00
− 0.03 −0.09 −0.11
0.07 −0.05 0.08 − 0.05 −0.13
1.00
− 0.07 0.00 0.06 −0.07 0.05 − 0.02 − 0.12
0.19 −0.26 1.00
0.05 0.01 0.01 −0.07 0.06 − 0.03 0.01 0.00 −0.16 −0.20
0.08 0.01 −0.01 −0.04
0.02 −0.01 0.00 0.01
0.05 −0.01 0.09 0.01
0.08
0.04
0.12
0.07
0.04 − 0.09
1.00
0.63
0.29
1.00
0.20
0.78
0.60
0.09
1.00
0.09
0.72
0.08
0.48
0.26
0.26
0.24
−0.22
−0.19
0.16
0.09
−0.26
−0.09
0.05
0.02
0.08
0.52
0.28
0.40
0.16
0.43
0.15
−0.19
−0.05
0.10
0.04
−0.12
0.03
−0.04
0.00
− 0.06
0.18
0.10
0.35
0.04
0.09
0.19
− 0.10
− 0.06
− 0.01
− 0.01
0.05
0.03
0.01
− 0.09
0.19
0.58
0.26
0.38
0.25
0.35
0.10
− 0.07
− 0.06
0.08
0.11
− 0.13
− 0.03
− 0.08
0.11
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
1.00
0.10 −0.02 − 0.03
0.05 0.03 − 0.05
0.11
0.02 0.13
−0.12
10
0.03 − 0.01
0.03
0.05
0.02
0.22 0.22 − 0.04 0.22 0.09 0.20 0.23 − 0.02
0.20 0.21 − 0.04 0.15
0.06 0.19
0.22 0.05
0.39 0.33 0.15
0.09 0.03 0.20 −0.03 0.17
0.25 0.46 0.04 0.37 0.16
0.25 0.32 0.18
−0.21 −0.23 0.10 − 0.24 −0.31 −0.20 −0.08 − 0.17
0.13
0.26 0.09 0.11
0.22 0.14 −0.02 0.28
0.23 0.11 –0.13
0.24 0.29 0.20 0.14
0.01
0.09 0.15 − 0.08 0.09 0.15
0.12
0.19 − 0.01
0.05 −0.10
0.07
0.02 0.03 −0.03
0.07 0.13
0.03
0.07
0.05
− 0.01
0.05
− 0.03
− 0.06
0.03
− 0.01
0.02 −0.05 − 0.09
0.03 0.00 0.07
0.02 −0.01 0.03
0.09 −0.03 −0.04
0.16
0.11
0.08
0.34
0.06
0.00
0.01
0.01
−0.12
0.00
0.12
−0.11
−0.03
0.03
0.02
−0.05
0.00
1.00
0.13 −0.04 −0.18 −0.23 −0.14
1.00
0.02 −0.03 −0.16 −0.20 −0.12 − 0.14
1.00
0.03 −0.02 −0.14 −0.18 −0.11 − 0.13 − 0.11
1.00
0.03 − 0.01 −0.04 −0.12 −0.15 −0.09 − 0.10 − 0.09 −0.08
−0.02 0.15 − 0.02 0.00 0.03 −0.06 −0.06 0.10
−0.07 0.06 0.07 0.08 −0.12 −0.08 0.05 −0.06
0.01 −0.20 0.12 − 0.10 −0.06 0.06 −0.01 0.03
−0.15
0.10
0.12 − 0.11
0.00 0.08 −0.12
0.01
0.08 0.00 0.01 − 0.12
0.15
0.03 −0.02 0.03
0.02 −0.10
0.06 0.08 −0.13
0.00 −0.05 −0.06
– 0.04 – 0.04 – 0.01 0.06 –0.12
0.05 0.02 0.04
−0.06 0.07 0.03 0.03 −0.10
0.06 0.04 −0.01
− 0.04
− 0.07
− 0.07
− 0.13
− 0.05
0.06
0.01
0.00
1.00
0.03 0.01 0.14
1.00
0.14
0.02 0.04 0.29 1.00
− 0.04 0.09 −0.08 − 0.07 0.09 1.00
0.05 0.08 0.07 0.21 0.22 0.08 1.00
− 0.06 −0.02 −0.08 − 0.07 −0.11 −0.06 −0.10
1.00
− 0.01 0.03 0.12 − 0.02 0.00 0.11
0.04 − 0.49 1.00
0.04 −0.04 −0.06 0.16
0.17
0.16
0.12 − 0.33 −0.38 1.00
− 0.09 −0.04 −0.04 0.09 0.15
0.01 0.06 − 0.05 −0.09 0.32 1.00
J.Y.C. Ho et al. / Intern. J. of Research in Marketing 26 (2009) 168–179
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
ln(Number of Theaters)
ln(Box Office Revenues)
ln(SB Advertising
Spending)
ln(Regular TV Advertising
Spending)
ln(Competition Intensity)
ln(Production Budget)
ln(Movie Runtime)
ln(Advertising Lead Time)
ln(Pre-Launch Buzz)
ln(Publicity)
Binary: GENRE—Action
Binary: GENRE—Comedy
Binary: GENRE—Drama
Binary: MPAA-PG Rated
Binary: MPAA-PG13 Rated
Binary: MPAA-R Rated
Binary: Distributor—Disney
Binary: Distributor—AOL
Binary: Distributor—
Viacom
Binary: Distributor—Sony
Binary: Distributor—Fox
Binary: Distributor—
Vivendi
Binary: Distributor—
DreamWorks
Binary: Star Power
Binary: Director Power
Binary: Sequel
ln(Critics Rating)
Binary: SEASON—Jan–Apr
Binary: SEASON—May–Aug
Binary: SEASON—Nov–Dec
Binary: Holidays
⁎ Highlighted numbers are significant at 5% level.
173
Author's personal copy
174
J.Y.C. Ho et al. / Intern. J. of Research in Marketing 26 (2009) 168–179
bars. Our operationalization of several variables departs from the
literature (e.g., BDT 2006). For example, we measure pre-launch buzz
as the amount of interest in the movie on the IMDb website, as
compared to using the number of theaters showing the movie or
revenues per screen as an indicator of post-consumption word-ofmouth. We also introduced several new control variables, such as the
lead time from the start of the movie's launch TV advertising
campaign to its release date (LEADj) and the pre-launch publicity
received (PUBj), as measured by coverage of the film in Entertainment
Weekly.
In Table 2, we present summary statistics for the continuous
variables for the Super Bowl and non-Super Bowl advertised movies.
As expected, Super Bowl movies have higher average box office
revenues and theater engagement numbers in the opening week as
compared to non-Super Bowl movies, even though our sample
consists of only wide-release movies. Super Bowl movies also have
larger launch TV advertising expenditures (TVADj), higher production
budgets (BUDGETj), a longer lead time from the first major TV
advertising effort to release week (LEADj) (as in our earlier discussion
of Fig. 1), longer runtime (RUNTIMEj), less competition (COMPj), a
higher level of pre-launch buzz (BUZZj), more publicity (PUBj), and
better critics' ratings (CRITICSj), suggesting that these variables are
potential confounding factors with the use of Super Bowl TV advertising. In terms of rating, Super Bowl movies tend to be PG13 or Rrated action movies released between February and August. Compared to non-Super Bowl movies, Super Bowl movies are more likely
to be sequels.11 Table 3 presents the correlations among the key
variables in the study. None of the correlation estimates among our
control variables are high in terms of magnitude. To test whether
multicollinearity is an important factor in our results, we calculated
the Variance Inflation Factor (VIF) for the variables in Eqs. (1) and (2).
A standard rule of thumb (Belsley, Kuh, & Welsch, 1980) is that the VIF
be below 10; we obtained average estimates of 3.04 and 3.62,
respectively, and none of the variables had a VIF exceeding 10.
5. Model estimation
As discussed earlier, our estimation procedure must address the
complex nature of the endogenous variables in our model. Specifically,
Super Bowl advertising decisions can potentially be endogenous. To
overcome this problem, we first estimate a probit model to approximate
the Super Bowl advertising decision using a set of exogenous variables
(Stage 1). By multiplying the probit's predicted probability of advertising
in the Super Bowl by the unit cost of Super Bowl advertising expenditure
(i.e., the cost of a 30-second advertisement), we obtain the expected
Super Bowl expenditure for each movie. We then substitute this
estimated expenditure for SUPERBOWLj when estimating Eqs. (1) and
(2) (Stage 2). These predicted or expected values only contain the part of
the variation in SUPERBOWLj that is due to the exogenous variables,
which are uncorrelated with the errors in Eqs. (1) and (2); thus, the
potential endogeneity due to unobserved characteristics of SUPERBOWLj
is removed from the model.12 As shown in Mroz (1999), two-stage
procedures such as ours will result in consistent and asymptotically
unbiased estimates. To validate our results, we employ several tests to
check for the consistency and appropriateness of our instruments to
estimate the model. Unlike BDT (2006), we do not specify an equation for
total regular launch TV advertising expenditure. Instead, we control for
the endogeneity of regular TV advertising TVADj by the use of the
11
Details of Super Bowl and non-Super Bowl movie characteristics are available from
the authors upon request.
12
Heckman (1978) first proposed a similar approach to control for discrete
endogenous variables. In labor economics, a similar approach has been widely used
(for example: Lee, 1978; Blundell & Powell, 2004; Sandy & Elliot, 1996). In an
unpublished working paper in marketing that we were not aware of when we
undertook this research, Luan and Sudhir (2007) used a similar approach to control for
advertising endogeneity.
instrumental variable method. We abstract away from specifying
additional equations for the movie distributors'/studios' advertising
decision processes in order to keep the model parsimonious and tractable, and also to focus on the two critical outcomes of theater
engagements and box office revenues. This approach also helps reduce
unintentional bias from directly estimating the advertising decision
process. A similar approach has been used in empirical demand analysis
to avoid obtaining biased parameter estimates in new empirical
industrial organization studies (Dube & Chintagunta, 2003).
The variables used in the probit model are BUDGETj, GENREj,
MPAAj, DISTRIBUTORj, SEQj, HOLIDAYj and the time difference in weeks
between the Super Bowl week and the week preceding the release of a
movie (denoted as SB_DISTANCEj):
ð3Þ
ProbðAdvertising in Super Bowl by movie jÞ
= Φðα0 + α1 ⋅BUDGETj +
2
α2 ⋅BUDGETj
+ α3 ⋅SB DISTANCEj
+ α4 ⋅HOLIDAYj + α5 ⋅SEQj + ∑ αk ⋅GENREjk + ∑ αm ⋅MPAAjm
+ ∑ αn ⋅DISTRIBUTORjn Þ
k
m
n
where Φ(.) is the cumulative distribution function of the standard
normal distribution.13
Super Bowl movies are neither the most expensive nor the least
expensive of those released by the distributors/studios. This implies
that there can be nonlinear effects of the production budget on Super
Bowl advertising decisions. To control for such a non-linear effect, we
added squared BUDGETj as an explanatory variable. In addition, as no
Super Bowl movies in our sample were G-rated or distributed by
Viacom, we dropped the corresponding dummy variables in our estimation. To avoid perfect multicollinearity, we also dropped one
binary of each set of the categorical variables (specifically, the
indicators of family movies and movies distributed by small
distributors).
Table 4 presents the probit regression results. Of the variables used
to specify the probit model, BUDGETj, BUDGET2j , SB_DISTANCEj, and the
distributor binaries for Disney and Vivendi are significant at p b 0.05.
The significance of the budget variables suggests that there is a nonlinear effect of the production budget on the Super Bowl advertising
decision. Up to a level of $49 million, the higher the production budget, the more likely a movie is to be advertised during the Super Bowl,
but after that the likelihood declines. As expected, the negative and
significant SB_DISTANCE suggests that the longer the time between
the Super Bowl and the movie release date, the less likely is Super
Bowl advertising. The significance of Disney and Vivendi suggests that
distributors can have specific and significantly different strategies in
terms of Super Bowl advertising.
There are three other endogenous variables on the right hand side
of the model (i.e., THEATERj in Eq. (2) and TVADj in both equations).
We control for their endogeneity by two instrumental variables in
addition to the exogenous variables already included in Eqs. (1) and
(2). They are [1] BUZZt − 1, j: the buzz in the week before the release
week, and [2] TVADt − 1, j: the cumulative regular advertising expenditure in the week before the release week. Specifically, in the theater
engagement Eq. (1), to control for the endogeneity of TVADj, we add
both TVADt − 1,j and BUZZt − 1, j (i.e., ln(TVADt − 1, j) and ln(BUZZt − 1, j) in
the linearized Eqs. (1) and (2)). In the box office revenue Eq. (2), to
control for the endogeneity of THEATERj and TVADj, besides using
TVADt − 1, j and BUZZt − 1, j, we also use the instrumental variables
RUNTIMEj, BUDGETj and DISTRIBUTORj, which are excluded from
Eq. (2) as explanatory variables.
13
As conglomerates own both movie studios and television networks (e.g., Disney
and ABC are jointly owned), we tested to see if movie studios were more likely to
advertise when the Super Bowl was shown on their jointly owned networks, but found
no such effects.
Author's personal copy
J.Y.C. Ho et al. / Intern. J. of Research in Marketing 26 (2009) 168–179
Table 4
Parameter estimates of the Probit Model of Super Bowl advertising.
Variables
Parameter estimates
Intercept
Budget
Budget2
Distance between SB & release
Movie runtime
Binary: holidays
Binary: Sequel
Binary: Genre—Action
Binary: Genre—Comedy
Binary: Genre—Drama
Binary: Distributor—Disney
Binary: Distributor—AOL/TW
Binary: Distributor—Sony
Binary: Distributor—Fox
Binary: Distributor—Vivendi
Binary: Distributor—Dream Works
Binary: MPAA-PG13
Binary: MPAA-R
Pseudo R2
− 6.189
0.098
− 0.001
− 0.058
0.015
− 0.437
0.711
− 0.410
− 0.850
− 1.157
1.312
0.635
0.876
0.432
1.020
0.308
0.699
0.938
0.653
Highlighted and underlined numbers are significant at 5% level. Model fit measure:
McKelvey and Zavoina's pseudo R2 = 0.653.
To demonstrate that our approach controls for the endogeneity and
provides consistent estimates, we use a set of diagnostic tests. We first
test their endogeneity by the Hausman test (Greene, 2003). Similar
to BDT (2006), we find that in the second stage estimation, total
regular launch TV advertising and theater engagements are endogenous (test statistic = 46.72 with p-value b 0.001). We also use a test of
overidentification to examine the appropriateness of our instruments
(Woolridge, 2002) and do not reject the null hypothesis, implying that
our sets of instruments are orthogonal (i.e., not correlated) with the
error structures of the model (test statistic = 6.77, p-value = 0.75).
In addition to examining the statistical significance of the effect of
theater engagement in the box office Eq. (2), we also demonstrate the
mediating role of distribution coverage in the Super Bowl advertising
effect on the opening week box office performance by comparing our
dual path model (DP model) with an alternative model without such
mediation by distribution coverage. Specifically, we drop the theater
variable from the box office equation, creating a non-mediated model
(NM model). We use a likelihood ratio test to compare these two
models under the assumptions of normally distributed errors (Greene,
2003, page 409). We reject the null hypothesis that there is no
significant difference between the two models (LR test statistic = 4.27,
p b 0.01). This implies that the unrestricted model (DP) is preferable to
the restricted model (NM). To compare these two models further, we
estimated the system R2 (McElroy, 1977). By this measure, the DP
model fits substantially better than the NM model with a system
R2 = 0.701 compared to 0.59. These diagnostics suggest that having
the theater variable playing a mediating role in the DP model
significantly increases the explanatory power of the DP model over
the NM model.
6. Results and discussion
Columns A & B in Table 5 present the parameter estimates of the DP
and NM models. We use GMM estimation to estimate Eqs. (1) and (2)
simultaneously. Woolridge (2002) terms this approach GMM-3SLS.
BDT and EE used standard 3SLS estimation, which is a restricted
version of the procedure we use. Compared to standard 3SLS, the
GMM-3SLS approach allows us to specify equation-specific sets of
instruments, thereby providing flexibility in estimation.14 Greene
14
Woolridge (2002) (pages 196–198) provides a detailed discussion of the key
differences between GMM-3SLS and standard 3SLS techniques in estimating a system
of equations. We also estimated the model using 2SLS, and the estimates are very
similar to the estimates from 3SLS.
175
(2003) states that in the presence of unknown heteroskedasticity,
system of equations estimation using GMM is also more efficient. For
the purpose of comparison, we also present results for our DP model
using the standard 3SLS technique in Table 5 (column C). As the table
shows, the parameter estimates for our focal variables, THEATERj,
TVADj, and SUPERBOWLj, are qualitatively similar across the two
different estimation methods. In terms of system R2, the GMM-3SLS
model fits slightly better than the 3SLS model. Given the flexibility and
efficiency of the GMM-3SLS approach, we use the GMM 3SLS results
for the rest of the analysis.
One binary variable of each set of categorical variables, GENRE,
MPAA, DISTRIBUTOR and SEASON, is dropped to avoid perfect multicollinearity in the estimation process. In particular, we drop the binary
variables that indicate: [1] the movie is a family movie, [2] it is rated G
by the MPAA, [3] it is distributed by one of the smaller distributors,
and [4] it is released in the September–October season. When interpreting the effects of binary variables, such as whether a movie is Rrated, we should note that the estimated parameter associated with
MPAA-R captures the effects of MPAA-R relative to the base case, a
G-rated family movie released in September–October (by a small
distributor if in the theater equation). The highlighted parameter
estimates in Table 5 are significant at the 5% level, and t-test statistics
are generated using robust standard errors. For comparison purposes
we also estimated a model without any control for endogeneity.15 The
values of the parameter estimates and significance levels change quite
dramatically once we control for endogeneity; more parameters become significant once we control for endogeneity, suggesting an
improvement in statistical efficiency (i.e., a decrease in estimated
standard errors). This is not unusual as Villas-Boas and Winer (1999)
showed that controlling for endogeneity improves the efficiency of
parameter estimates.
6.1. Hypothesis testing
In H1a, we hypothesized that Super Bowl advertising spending would
have a positive direct effect on opening week box office revenue, even
when the mediating role of distribution coverage was controlled for. As
β2,SB is not significantly different from zero at p b 0.05, we cannot find
support for this hypothesis. This result is in contrast to Yelkur et al.'s
(2004) exploratory study, which did not control for the endogeneity of
Super Bowl advertising and the mediating effect of theaters.
In H1b, we hypothesized that movie exhibitors play a mediating
role in the causal chain from Super Bowl advertising to opening week
box office revenues. We test this by examining the two parts of the
indirect path. The estimate for β1,SB in the theater equation suggests a
significant positive effect of Super Bowl advertising spending on
number of theaters, establishing the first part of the indirect path. The
second part of the indirect path, the estimate for β2,THR in the box
office equation, is also positive. Consequently, our estimation results
support this hypothesis. Similar to Super Bowl advertising spending,
regular TV advertising spending in the pre-launch period influences
initial box office revenues mainly through the theater engagement
factor. In particular, we found that TVADj's effect on BOj works only
through THEATERj (significant positive β1,AD but insignificant β2,AD).
Our second hypothesis, H2, requires us to compare the effectiveness of Super Bowl and regular TV advertising expenditures. Given the
log–log specification, the estimated parameters are also the elasticity
estimates. Thus, the elasticity of theater engagements with respect to
Super Bowl expenditures is 0.012 and with respect to regular TV
advertising expenditures is 0.306. Both of these estimates are
statistically significant. On the other hand, in the box office equation,
the elasticities of box office revenue with respect to Super Bowl
advertising and with respect to regular TV advertising are both
15
Detailed results for this model are available from the authors upon request.
Author's personal copy
176
J.Y.C. Ho et al. / Intern. J. of Research in Marketing 26 (2009) 168–179
Table 5
Parameter estimates of the regression models.
Model
(A)
(B)
Variables
Dual Path Model (DP)-GMM
Non-Mediated Model (NM)-GMM
Dual Path Model (DP)-3SLS
Theater estimates
Theater estimates
Theater estimates
Intercept
ln(Production Budget)
ln(SB Advertising Spending)
ln(Regular TV Advertising Spending)
ln(Number of Theaters)
ln(Critics Rating)
ln(Competition Intensity)
ln(Movie Runtime)
ln(Pre-launch Buzz)
ln(Advertising Lead Time)
ln(Publicity)
Binary: Sequel
Binary: Genre—Action
Binary: Genre—Comedy
Binary: Genre—Drama
Binary: MPAA-PG Rated
Binary: MPAA-PG13 Rated
Binary: MPAA-R Rated
Binary: Distributor—Disney
Binary: Distributor—AOL
Binary: Distributor—Viacom
Binary: Distributor—Sony
Binary: Distributor—Fox
Binary: Distributor—Vivendi
Binary: Distributor—DreamWorks
Binary: Star Power
Binary: Director Power
Binary: Season Jan–Apr
Binary: Season—May–Aug
Binary: Season–Nov–Dec
Binary: Holidays
Over identification test
System-R2
3.371
0.050
0.012
0.306
− 0.030
− 0.226
0.016
0.037
0.000
0.142
− 0.050
− 0.111
− 0.059
− 0.080
− 0.110
− 0.168
− 0.160
− 0.050
− 0.092
− 0.070
− 0.060
− 0.136
− 0.206
0.075
0.040
Box office estimates
5.860
0.008
0.149
0.933
0.891
− 0.164
0.066
0.088
0.021
0.506
0.244
0.086
0.180
− 0.065
− 0.018
− 0.115
0.165
− 0.065
− 0.005
0.406
0.220
0.258
6.77 (df 10), p-value: 0.75
0.701
(C)
3.375
0.045
0.013
0.303
− 0.026
− 0.206
0.013
0.038
− 0.001
0.142
− 0.040
− 0.097
− 0.054
− 0.063
− 0.094
− 0.152
− 0.146
− 0.050
− 0.081
− 0.064
− 0.060
− 0.130
− 0.151
0.074
0.039
Box office estimates
8.534
0.018
0.457
0.748
− 0.175
0.073
0.128
0.020
0.656
0.142
− 0.076
0.041
− 0.070
− 0.047
− 0.189
0.241
0.008
− 0.028
0.378
0.147
0.289
7.30 (df 11), p-value: 0.77
0.59
3.526
0.040
0.015
0.297
− 0.038
− 0.175
0.017
0.033
− 0.001
0.122
− 0.065
− 0.102
− 0.076
− 0.071
− 0.113
− 0.148
− 0.164
− 0.043
− 0.062
− 0.068
− 0.061
− 0.115
− 0.200
0.054
0.051
Box office estimates
6.253
0.004
0.107
0.968
0.985
− 0.216
0.071
0.080
0.019
0.510
0.261
0.096
0.152
− 0.047
0.053
− 0.007
0.168
− 0.090
0.050
0.458
0.331
0.234
0.69
Highlighted and underlined numbers are significant at 5% level.
insignificant. Nevertheless, the magnitude of the coefficients for TV
advertising is greater than those for Super Bowl advertising. This
suggests that the first dollar available for advertising should be spent
on regular TV advertising. However, the average movie in our sample
spends $13 million in TV advertising, while most movies spend zero in
Super Bowl advertising. Given the diminishing marginal returns from
all advertising (all elasticities are smaller than one), initial Super Bowl
advertising can provide larger returns for movies than spending at or
above the average expenditure level. To explore this issue further, we
conducted a counterfactual simulation analysis.
6.2. Counterfactual simulation
We conducted a counterfactual simulation by giving each movie in
our sample an additional $2.2 M, which is equal to the average rate for
one 30-second spot during the Super Bowls of 2000–2002, to spend
on either Super Bowl or other TV advertising opportunities. We
created the following three scenarios:
Scenario 0 (Base Case): Each movie does not spend the additional
$2.2 M on either Super Bowl or other TV advertising. (This is
used as the benchmark for comparison).
Scenario 1 (with extra Super Bowl): Each movie spends its
additional $2.2 M on Super Bowl advertising.
Scenario 2 (with extra AD): Each movie spends its additional $2.2 M
on regular TV advertising.
Fig. 3a shows the simulated values of theater engagement (THEATERj )
and box office revenues (BO j) for each of the three scenarios. The average
of the simulated values of theater engagement and box office revenues
shows that Scenario 1 provides higher returns than Scenario 2, i.e., the
marginal effect of Super Bowl advertising on the first week box office
revenues is larger than the marginal effect of regular TV advertising in
most cases. To show this more clearly, we subdivide our results into those
movies that did and did not actually advertise during the Super Bowl
(Fig. 3b). Spending an additional $2.2 M on regular TV advertising by nonSuper Bowl movies increases first week box office revenues from $23.89 M
to $25.50 M; however, spending the same amount on Super Bowl
advertising increases first week box office revenues to $31.37 M. On the
other hand, a $2.2 M extra expenditure by a Super Bowl movie on Super
Bowl advertising generates $43.35 M in revenue, an increase of only
$0.54 M from the base case. If a Super Bowl movie had spent the same
amount of money on regular advertising, then the increase in first week
box office revenues would have been an average of $2.84 M above the base
case. This simulation exercise suggests that Super Bowl movies should
only buy one Super Bowl advertisement, which is the usual practice for
Super Bowl movies.
6.3. Additional results of interest
Whenever directional hypotheses are possible, our significant
(p b 0.05) results are always in the expected direction. We now briefly
summarize our key significant results for the other key variables.16
16
We do not discuss the buzz variable here as it is insignificant in the GMM
estimation. However, as it is significant in the 3SLS estimation, there is a need for
further research on this interesting variable.
Author's personal copy
J.Y.C. Ho et al. / Intern. J. of Research in Marketing 26 (2009) 168–179
177
Fig. 3. a: Simulation results under three scenarios. b: Simulation results under three scenarios by Super Bowl (SB) and non-Super Bowl (NSB) movies.
1) Publicity (PUBj): The significant positive effect of publicity on first
week box office revenues suggests that moviegoers are influenced
by the amount of media attention surrounding individual movies.
On the other hand, there is no significant effect of publicity on
theaters, suggesting that theater managers are less influenced by
publicity. Although publicity is a new measure in the movie
literature, Hollywood studios spend considerable sums of money
to generate publicity, and its further study would be valuable.
2) Sequel (SEQj): The sequel binary is positively significant in both the
theater and box office equations (at the 5% level). In other words,
sequels appear to have built-in brand equity, making them
attractive to movie exhibitors and moviegoers.
3) Lead time (LEADj): The lead time between the major TV campaign
start date and the movie release date has a significant positive
effect on theaters, but no direct effect on box office.
4) Critics' Review (CRITICSj): The significant positive effect of critics'
ratings on opening week box office revenue is consistent with the
recent results in Basuroy, Chatterjee, & Ravid (2003).
5) Star Power (S_POWERj): The significant positive effect of star
power on the number of theaters showing the movie in the first
week but not on first week box office suggests that star power
mainly influences opening week box office indirectly through
increasing distribution coverage.
6) Seasons (SEASONj) and Holidays (HOLIDAYj): As expected, movies
released in the summer season of May–August, in the holiday
season of November–December, and during a week including a
major holiday fared better in opening week box office revenue
than movies released at other times of the year.
7. General discussion
In this paper we explore the role of major event advertising on
market outcomes using advertising data from the movie industry.
Utilizing a sample of movies from the years 2000–2002, this paper
demonstrates that both Super Bowl and regular TV advertising
influence opening week box office revenue, primarily through
increasing the number of movie exhibitors showing the movie in its
first week of wide release. While these results may be specific to the
movie market, they suggest that failure to take into account
Author's personal copy
178
J.Y.C. Ho et al. / Intern. J. of Research in Marketing 26 (2009) 168–179
distribution in predicting the influence of advertising on new product
demand can be problematic.
To the best of our knowledge this is the first paper to differentiate
between major event and regular TV advertising in econometrically
estimating their comparative effects. In cluttered and fragmented
media markets, major event advertising can play a significant role
for products targeting a national audience and pursuing wide
distribution. Our analysis suggests that major event advertising can
play a distinctive role in building distribution and demand.
We find that Super Bowl advertising is not as effective as regular
TV advertising when both are evaluated at the same level. In part, the
comparatively lower effectiveness of Super Bowl advertising may
occur because it is a one-time occurrence that is typically distant from
the actual showing of the movie; by contrast, regular TV advertising
runs over a number of weeks and typically occurs closer to the movie's
opening. However, Super Bowl advertising is still attractive in many
cases, as shown in our counterfactual analysis. It can usefully provide
better returns after a minimum threshold of regular TV advertising is
reached. This result suggests that firms with limited budgets may be
better off using regular television advertising rather than Super Bowl
advertising. While our results indicate that the use of Super Bowl
advertising for at least some movies with above average advertising
budgets may be more profitable than increased spending on regular
advertising, we find little support for a movie to invest in more than
one Super Bowl advertisement. In terms of methodology, in this paper
we apply a two-stage instrument variable estimation technique to
estimate a model with discrete endogenous explanatory variables. In
the first step we model the discrete choice process, and in the second
step we estimate a system of theater and box office revenue equations
after incorporating the estimated probability of advertising in the
Super Bowl from the first step. This approach helps us to generate
consistent and asymptotically unbiased parameter estimates.
At the same time, we acknowledge several limitations in this paper
and believe these limitations open interesting future research
opportunities.
First, it would be interesting to examine the effects of Super Bowl
advertising in other product categories. Our results provide insight for
marketers seeking to introduce new products and build distribution.
For many other categories, Super Bowl advertising is part of an
ongoing advertising campaign for products already in the market.
Thus, it becomes a considerable econometric challenge to separate out
the effects of other advertising and continuing sales in order to
measure the impact of Super Bowl advertising. On the other hand,
such product categories also provide the opportunity to study the
effect of multiple ads for the same brand during one Super Bowl.
Second, our research indicates the critical role of advertising in
influencing marketing channel intermediaries. To understand the
overall impact of advertising, we need to model the influence of
advertising on the channel intermediaries and on other stakeholders
in the production and marketing processes. For example, in recent
years, a number of retailers and airlines have featured their own
employees in their advertising campaigns. Does such advertising first
influence productivity (by improving employee morale) and then
directly or indirectly improves sales?
Many CEOs are questioning the efficacy of large scale advertising
spending, especially in this age of fragmented media markets where
consumers also control the content to which they are exposed through
devices like TiVo. The marketing metrics literature is one response to
this call. Further research that quantifies and provides detailed
pathways of effects for both major event advertising and standard
advertising expenditures is, in our view, extremely valuable.
Appendix A. Derivation of buzz variable
We measure pre-launch buzz for a specific movie based on the
number of page views for that movie on IMDb.com in its release week.
Specifically, IMDb.com reports, for each of the movies in its database, a
weekly measure called MOVIEmeter. MOVIEmeter is a rank measure of
individual movies' total page views in a specific week that we convert,
using an inverse function, to a measure of share of views. To allow a
comparison of two equally ranked movies in two different time
periods (say, one movie ranked second in August 2001 and another
movie with a second-place MOVIEmeter rank in March 2002), we use
additional assumptions and data sources to convert the rank measure
of MOVIEmeter into a traffic-based measure of pre-launch buzz for
movie j in its release week t, BUZZjt. Specifically, we use the following
function:
"
#
1
BUZZjt =
*IMDb―Traffict
MM―Rankjt + C
ðA1Þ
where MM_Rankjt = MOVIEmeter rank for movie j in its release week t
and IMDb_TRAFFICt is the estimated number of visits to IMDb.com in
movie j's release week t.
h
i
1
, is our assumed
The first part of expression (A1), MM Rank
+ C
―
jt
relation between the observed rank of a movie and its share of total
visitors, which is unobserved to us. C is a constant and in estimation, we
h
i
1
, over the top
set C at 60 such that the sum of the share, MM Rank
+
C
jt
―
100 ranked movies in a given week does not exceed one.
The second part of expression (A1) is our estimate for the total
number of web visits to IMDb.com in a specific week. While we could
not observe the actual number of visits to IMDb.com, we obtained
from Alexa.com the daily number of visitors to IMDb.com per one
million Alexa tool bar users from September 2001 to August 2004.
Assuming the proportion of Alexa tool bar users to the total Internet
user population remained the same, we estimated a regression model
capturing the seasonality of Alexa's IMDb.com traffic and then used
the regression model to extrapolate backwards the weekly traffic to
IMDb.com (i.e., IMDb_TRAFFICt), for the time frame of our data sample.
References
Ashley, R. C., Granger, W. J., & Schmalensee, R. (1980). Advertising and aggregate consumption: An analysis of causality. Econometrica, 48(5), 1149−1168.
Basuroy, S., Chatterjee, S., & Ravid, S. A. (2003). How critical are critical reviews? The
box office effects of film critics, Star Power and budgets. Journal of Marketing, 67(4),
103−117.
Basuroy, S., Desai, K. K., & Talukdar, D. (2006). An empirical investigation of signaling in
the motion picture industry. Journal of Marketing Research, 43, 297—295.
Belsley, D. A., Kuh, E., & Welsch, R. E. (1980). Regression diagnostics: Identifying influential
data and sources of collinearity. New York: Wiley.
Blundell, R. W., & Powell, J. L. (2004). Endogeneity in semi parametric binary response
models. Review of Economic Studies, 71(3), 655−679.
Chu, W. (1992). Demand signaling and screening in channels of distribution. Marketing
Science, 11(4), 327−347.
Desai, P. S. (2000). Multiple messages to retain retailers: Signaling new product
demand. Marketing Science, 19(4), 381−389.
Dube, J., & Chintagunta, P. (2003). Comment. Quantitative marketing and economics, 1,
293−298.
Einav, L. (2007). Seasonality in the U.S. motion picture industry. Rand Journal of Economics,
31(1), 127−145.
Elberse, A., & Eliashberg, J. (2003). Demand and supply dynamics for sequentially released
products in international markets: The case of motion pictures. Marketing Science, 22
(3), 329−354.
Greene, W. H. (2003). Econometric analysis, 5th ed.: Prentice Hall.
Heckman, J. J. (1978). Dummy endogenous variables in a simultaneous equation system.
Econometrica, 46(4), 931−959.
Heyse, J., & Wei, W. W. S. (1985). Modeling the advertising–sales relationship through
use of multiple time series techniques. Journal of Forecasting, 4(2), 165−181.
Jones, J. M., & Ritz, C. J. (1991). Incorporating distribution into new product diffusion
models. International Journal of Research in Marketing, 8, 91−112.
Kanner, B. (2004). The Super Bowl of advertising: How the commercials won the game.
Princeton: Bloomberg Press.
Lee, L. F. (1978). Unionism and wage rates: Simultaneous equations model with qualitative
and limited dependent variables. International Economic Review, 19(2), 415−433.
Leenheer, J., Van Heerde, H. J., Bijmolt, T. H. A., & Smidts, A. (2007). Do loyalty programs
really enhance behavioral loyalty? An empirical analysis accounting for selfselecting members. International Journal of Research in Marketing, 24, 31−47.
Author's personal copy
J.Y.C. Ho et al. / Intern. J. of Research in Marketing 26 (2009) 168–179
Luan, Y. J. & Sudhir, K. (2007). Forecasting Advertising Responsiveness for Short LifeCycle Products. Working paper, Tuck School of Business at Dartmouth.
Mayzlin, D. (2006). Promotional chat on the Internet. Marketing Science, 25(2), 155−163.
McElroy, M. (1977). Goodness of fit for seemingly unrelated regressions: Glahn's R2y, x
and Hooper's r2. Journal of Econometrics, 6(3), 381−387.
Montgomery, D. B. (1975). New product distribution: An analysis of supermarket buyer
decisions. Journal of Marketing Research, 12, 255−264.
Mroz, T. A. (1999). Discrete factor approximations in simultaneous equation models:
Estimating the impact of a dummy endogenous variable on a continuous outcome.
Journal of Econometrics, 92, 233−274.
Newell, S. J., & Henderson, K. V. (1998). Super Bowl advertising: Field testing the importance
of advertisement frequency, length and placement on recall. Journal of Marketing
Communications, 4(4), 237−248.
Newell, S. J., Henderson, K. V., & Wu, B. T. (2001). The effects of pleasure and arousal on
recall of advertisements during the Super Bowl. Psychology and Marketing, 18(11),
1135−1153.
Olver, J. M., & Farris, P. W. (1989). Push and pull: A one-two punch for packaged products.
Sloan Management Review, 31(1), 53−61.
Pavelchak, M. A., Antil, J. H., & Munch, J. M. (1988). The Super Bowl: An investigation
into the relationship among program context, emotional experience, and ad recall.
Journal of Consumer Research, 15(3), 360−367.
179
Russell, C. A. (2002). Investigating the effectiveness of product placements in television
shows: The role of modality and plot connection congruence on brand memory and
attitude. Journal of Consumer Research, 29(3), 306−318.
Sandy, R., & Elliot, R. F. (1996). Unions and risk: Their impact on the level of compensation
for fatal risk. Economica, 63(250), 291−309.
Tomkovick, C., Yelkur, R., & Christians, L. (2001). The USA's biggest marketing event
keeps getting bigger: An in-depth look at Super Bowl advertising in the 1990s.
Journal of Marketing Communications, 7(2), 89−108.
Vakratsas, D., & Ambler, T. (1999). How advertising works: What do we really know?
Journal of Marketing, 63(1), 26−43.
Villas-Boas, J. M., & Winer, R. (1999). Endogeneity in brand choice model. Management
Science, 45, 1324−1338.
Woolridge, J. M. (2002). Econometric analysis of cross section and panel data. Cambridge,
MA: MIT Press.
Yelkur, R., Tomkovick, C., & Traczyk, P. (2004). Super Bowl advertising effectiveness:
Hollywood finds the games golden. Journal of Advertising Research, 44(1), 143−159.
“Great value at $80,000 per second,” Advertising Age, Jan 31, 2005.
“A February in Overdrive: Super Bowl and Olympics,” New York Times, Jan 18, 2006.