Author's personal copy - UBC Sauder School of Business · thousand viewers (CPM) of $19.85,...

This article appeared in a journal published by Elsevier. The attachedcopy is furnished to the author for internal non-commercial researchand education use, including for instruction at the authors institution

and sharing with colleagues.

Other uses, including reproduction and distribution, or selling orlicensing copies, or posting to personal, institutional or third party

websites are prohibited.

In most cases authors are permitted to post their version of thearticle (e.g. in Word or Tex form) to their personal website orinstitutional repository. Authors requiring further information

regarding Elsevier’s archiving and manuscript policies areencouraged to visit:

http://www.elsevier.com/copyright

http://www.elsevier.com/copyright

Author's personal copy

Playoff payoff: Super Bowl advertising for movies☆

Jason Y.C. Ho a,⁎, Tirtha Dhar b, Charles B. Weinberg b

a Simon Fraser University, 8888 University Drive, Burnaby, B.C., Canada V5A 1S6b Sauder School of Business, University of British Columbia, 2053 Main Mall, Vancouver, B.C., Canada V6T 1Z2

a b s t r a c ta r t i c l e i n f o

Article history:First received in 28 July 2008and was under review for 6 months

Area Editor: Richard Staelin

Keywords:Super BowlAdvertisingMoviesMarketing channel

Marketers are increasingly making use of major TV events, such as the Super Bowl, to advertise theirproducts. However, the economic value of such advertising is highly uncertain. Since an ad during the SuperBowl costs 2.5 times more per viewer reached than an ad during a network TV prime time show, developingmethods for evaluating such advertising and for measuring its effects seems particularly important. Using thesetting of the movie industry, this paper develops and estimates a model that includes both direct (onpotential moviegoers) and indirect effects (on exhibitors) of regular and Super Bowl advertising. The modelrecognizes the endogeneity of advertising, and in particular develops a discrete choice model to control forthe endogeneity of the Super Bowl advertising decision. The results indicate that Super Bowl advertising hasa 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; acounterfactual analysis, by contrast, shows that for a movie already spending at our sample's average TVspending level of $13 million, Super Bowl advertising has a greater effect on revenues than regular TVadvertising.

© 2009 Elsevier B.V. All rights reserved.

1. Introduction

Marketers, confronted on one hand by the increasing ability of con-sumers to avoid watching commercials aired during TV shows and onthe other by the sheer clutter of TV advertising, have turned both to“stealthmarketing,”where their presence is largely hidden, and tomajorTV event advertising, where their presence is written bold. While anincreasingnumberof papers look at the effects of stealthmarketing (e.g.,Russell, 2002; Mayzlin, 2006), little is known about major eventadvertising. Of all of themajor TVevents in theU.S., theannual broadcastof the Super Bowl (SB) is themost anticipated, discussed, and expensive.In this paper, our focus is on evaluating advertising during the SuperBowl, an iconic American eventwithmore than 90 million viewers eachyear — the most viewed TV show in the US. The second most viewedshow, the Academy Awards, attracts about half that number of viewers.To place a 30-second ad during the Super Bowl, marketers have to paymore than $2.3 million. Despite such high cost, neither industry(Advertising Age, Jan 31, 2005) nor academia can provide much insighton thevalue of advertisingduring theSuper Bowl inparticular andmajorTV events in general.

A Super Bowl advertisement is not only the most expensive spot onTV in absolute terms, but is also 2.5 times more expensive per viewerreached than regular advertising. In 2004, a 30-second prime timenetwork TV commercial cost approximately $120,500, with a cost perthousand 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 circum-stances underwhich a Super Bowl advertisement is a better investmentthan an advertisement aired during a regular TV show?

To answer this question, we examine the market impact of SuperBowl advertising for movies. Using data on the U.S. movie industry from2000 to 2002, we build a system of equations model to study thepotential effects of Super Bowl advertising on bothmovie exhibitors andmoviegoers. Our empirical results demonstrate that:

1. Super Bowl advertising for a movie influences the opening weekbox office revenues by indirectly attracting more movie exhibitorsto show the movie, thus increasing product availability, which inturn increases initial box office revenues. Super Bowl advertisingdoes not directly affect the moviegoers in the opening week (or insubsequent weeks).

2. Super Bowl advertising is not as effective as other TV advertisingexpenditures prior to the movie's release if both types of pre-launch TV advertising expenditures are evaluated at the sameinitial levels. On the other hand, given the presence of the well-documented diminishing returns to scale effect for advertising (seeVakratsas & Ambler (1999) for a review), we found that for a movie

Intern. J. of Research in Marketing 26 (2009) 168–179

☆ The authors gratefully acknowledge the support of the Social Sciences andHumanities 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

Contents lists available at ScienceDirect

Intern. J. of Research in Marketing

j ourna l homepage: www.e lsev ie r.com/ locate / i j resmar


that has already allocated the average amount of approximately$13 million that a mass market movie in our sample spends on TVadvertising, spending about $2.2 million on Super Bowl advertisingis usually more effective than adding the same amount of money tospending on regular TV advertising.

2. Related literature

Our goal in this paper is to use Super Bowl advertising by the movieindustry as a context inwhich to explore the unique role of major eventadvertising in amarketing channel after taking into account some of theunintentional shortcomings of earlier studies.

2.1. Major TV event advertising

Despite being an important marketing tactic, the value of major TVevent advertising has not received much academic attention. Withregard to the Super Bowl, the onlymajor 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 afield setting toexamine the effects of various design and media factors in advertising.An exploratory study by Yelkur, Tomkovick, and Traczyk (2004), alsofocusing on the movie industry, appears to be the sole exception.Compared to that study, we have developed a richer data set, an im-proved methodology and a more sophisticated model to describe howSuper Bowl advertising works through downstream channel membersto impact final consumer demand.

The lack of research on the value of major TV event advertising ispartly due to data challenges: in most major product categories, onlyone or two brands advertise during a specific major TV event. Inaddition, it is difficult to distinguish continuing sales from incremental

sales due to major event advertising. In light of these data challengesand research issues, we focus on the U.S. movie industry. In the threeyear period (2000–2002) that we examine, there were 19 differentmovies advertised during the three Super Bowl games.1 Only beer andsoft drink companies' brands had more Super Bowl advertisementsthan didmovies. Unlike these products, all of the advertisements in thecontext of our study were for yet-to-be released movies, thereby hel-ping us to avoid the seemingly intractable challenge of disentanglingthe effect of regular advertising on continuing sales from incrementalsales due to major event advertising.

2.2. Mediating role of downstream channelmembers in advertising effects

Although marketers primarily use advertising to stimulate con-sumers' demand (i.e., the “pull” effect), some advertising campaigns canalso have a “push” effect on retailers (Montgomery,1975; Olver & Farris,1989; Chu, 1992; Desai, 2000), and the increased product availability atthe retail level can then increase consumeradoption (Jones&Ritz,1991).Such “pull andpush” effects havebeennotedbypractitioners.2However,only a few empirical studies have formally examined such dual effects:

1 The percentage of observations possessing our focal characteristic, Super Bowladvertising, is similar to the percentage in previous studies in a similar context. Forexample, Basuroy et al. (2006) have about 6% of their sample being sequels, their focalvariable.

2 When Master Lock started its three-decade-long series of Super Bowl advertise-ments 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 ofCalifornia 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 weare here to stay.” (New York Times, Jan 18, 2006).

Fig. 1. Pre-launch advertising patterns of The Mummy Returns and U-571.

169J.Y.C. Ho et al. / Intern. J. of Research in Marketing 26 (2009) 168–179


Elberse&Eliashberg (2003; hereafter EE) andBasuroy, Desai, & Talukdar(2006); hereafter BDT) support the view that advertising positivelyaffects distribution intensity, which in turn increases final consumersales. Extending these two studies, which only consider aggregateadvertising spending, we decompose advertising spending into SuperBowl and regular advertising and compare the effectiveness of these twodifferent advertising tactics.

Fig. 1 illustrates the advertising patterns prior to the release weeksfor two movies, U-571 and The Mummy Returns. As can be seen, SuperBowl advertising tends to be separate from the overall pre-launchadvertising campaign; U-571's Super Bowl advertising appeared12 weeks before its release, and The Mummy Returns', 14 weeksbefore. The average start time for the Super Bowl advertised movies inour sample was 13 weeks before launch, as compared to 4 weeks forthe start of national advertising for non-Super Bowl advertisedmoviesin our sample. Thus, Super Bowl advertising not only involves largesums 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 advertisingand sales has been well documented in the advertising literature (e.g.,Ashley, Granger, & Schmalensee, 1980; Heyse &Wei, 1985). To measurethe true effect of advertising on sales, we thus need to control for theendogeneity of advertising. ExtendingEE (2003),whichonlymodels theendogeneity of distribution channels, BDT (2006) captures the en-dogeneity of both distribution channels and advertising. We furtherextend their studies by capturing the endogeneity of the Super Bowladvertisingdecisionprocess. This is an importantmethodological exten-sion because the Super Bowl advertising spending variable, from theperspective of a statistician, is similar to a binary choice variable; allSuper Bowl advertisedmovies have a spending level of about $2 million,while the non-Super Bowl movies have zero values. We test and correctfor the bias of the endogeneity of Super Bowl advertising by estimatingits effectwith a two-stage instrumentalvariable estimatormodified for adiscrete endogenous decision variable following Mroz (1999). Ourapproach is in contrast with other studies also controlling for a discreteendogenous variable, which usually use a linear regression model toapproximate the binary endogenous variable (e.g., Leenheer, VanHeerde, Bijmolt, & Smidts, 2007). In addition, as detailed below,extending EE (2003) and BDT (2006), we include advertising leadtime and directmeasures of pre-launchword-of-mouth and publicity tobetter control for the effects of these other communication variables onsales, thus further clarifying Super Bowl advertising's role in themarketing process.

3. Model development

To capture the potential “pull” and “push” effects of Super Bowladvertising for movies, we propose a dual path model. Fig. 2 depictsthe main model characterizing the relations among Super Bowladvertising, regular launch TV advertising, distribution coverage, andconsumer purchases. Consistent with EE (2003) and BDT (2006), wehypothesize that regular launch TV advertising has the same dualpaths as Super Bowl advertising does.3 For brevity, we hereafterdiscuss only the dual paths of Super Bowl advertising. There are twopaths by which Super Bowl advertising can influence consumer de-mand, namely a direct advertising effect on consumer demand and anindirect effect through downstream channel members.

We formalize the key features of our dual path model in the fol-lowing two hypotheses:

H1. Super Bowl advertising increases the opening week box officerevenues along two paths:

H1a (Direct Path). Super Bowl advertising directly increasesopening week box office revenues after controlling for the media-tion of the movie exhibitors.H1b (Indirect Path). Super Bowl advertising first increases thenumber of movie exhibitors. Then, the increased number of movieexhibitors increases opening week box office revenues.

H2. Super Bowl advertising is as effective as regular TV advertising.

Althoughwemodel onlyfirstweek sales to focus on the initial effectsof Super Bowl advertising, we also analyzed the effects of Super Bowladvertisingon subsequentweek sales.Qualitatively, the results for SuperBowl advertising do not change from those for the first week, as dis-cussed below.4

3.1. Econometric model

To test the key feature in our dual path model, we develop thefollowing system of equations model relating opening week box officerevenues for movie j (denoted as BOj), opening week number oftheaters engaged for movie j (denoted as THEATERj), regular launch

3 Essentially, we use TV advertising expenditure as a proxy for total advertisingexpenditure in the pre-launch period. According to 2003 MPAA market statistics, TVadvertising 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.

Fig. 2. Dual path model of Super Bowl TV advertising.

170 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 SuperBowl advertising expenditure by movie j (denoted as SUPERBOWLj):

THEATERj = eβ1:0 ⋅SUPERBOWLβ1;SB

j ⋅TVADβ1;AD

j

⋅∏∀kXβ1;k

Tkj ⋅e∑∀h

β1;h⋅ZThj⋅eε1

ð1Þ

BOj = eβ2:0 ⋅SUPERBOWLβ2;SB

j ⋅TVADβ2;AD

j ⋅THEATERβ2;THR

j

⋅∏∀kX

β2;k

Bkj ⋅e∑∀h

β2;h⋅ZBhj⋅eε2

ð2Þ

where X and Z are movie characteristics to be defined below, ε1 and ε2are the errors of the two equations, and the βs are the parameters to beestimated.

Unlike such previous studies as EE (2003) and BDT (2006), wedecompose total TV advertising spending into SuperBowl (SUPERBOWLj)and regular launch TV advertising expenditure (TVADj). Parametersβ2,SB

andβ2,AD in the boxoffice Eq. (2) capture the direct effects of Super Bowland 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 ofSuper Bowl and regular TV advertising spending by comparing theparameter estimates β2,SB versus β2,AD and β1,SB versus β1,AD.

To capture thewell-documented diminishingmarginal returns effectin advertising (Vakratsas & Ambler, 1999), both Eqs. (1) and (2) aremultiplicative, which allows either a concave (diminishing return toscale) or convex shape (increasing return to scale) for continuousvariables. 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 ofSuper Bowl and regular TV advertising at the same initial level, say bothbeing zero, and in a more realistic scenario, where an average moviespends approximately $13 million on regular advertising but zero onSuper Bowl advertising. Such a comparison, anticipating our empiricalresults, includes the effect of diminishing returns to scale of regular TVadvertising.

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 ofthe key explanatory variables in the theater equation. Specifically, BDT(2006) used realized total revenue as a proxy for expected totalrevenue in the first week theater equation. In this paper, we specifythe theater number as a function of a comprehensive set of availablecharacteristics in order to avoid some of the potential pitfalls of usingexpected revenue as an explanatory variable.5

XTkj and ZThj are the characteristics potentially influencing theatermanagers' screeningdecisions,with the former beinga set of continuousvariables and the latter being a set of indicator (binary) variables. Sim-ilarly, XBkj and ZBhj are the continuous and indicator variables potentiallyaffecting moviegoers' ticket purchase decisions. A majority of these Xkjand Zhj variables are common across Eqs. (1) and (2). These commonvariables are competitive intensity for movie j (COMPj), buzz (BUZZj),lead time of the launch advertising campaign (LEADj), amount ofpublicity generated (PUBj), movie j being a sequel (SEQj), movie j beingof a certain genre (GENREj), movie j's rating by MPAA (MPAAj), star

power (S_POWERj), and director power (D_POWERj). Table 1 providesdetailed definitions and information on the data sources.

Some variables are specific to each equation. Here we providereasons for their inclusion in each equation. The following three va-riables are unique to the theater equation (i.e., Eq. (1)): [1] Total pro-duction budget for movie j (BUDGETj): As theater managers are muchbetter informed about movie budgets than typical consumers, we useproduction budget in the theater Eq. (1), but not in the box officeEq. (2).6 [2] Run time of movie j (RUNTIMEj): Ceteris Paribus, theaterowners prefer shortermovies over longer ones because this allows them

5 We avoid using one single measure of expected total revenue for two reasons: [1] ifthe proxy for the expected box office revenue is highly correlated with theaters (inother words, theater owners are very good at predicting movie outcomes), then thesignificance of the variables of interest will diminish substantially; and [2] similarly, ifother explanatory variables influence the expected revenue (in other words, theatermanagers form their expectations based on available movie characteristics), then themodel will be overspecified, leading to a possible decrease in significance of the rest ofthe explanatory variables in the theater equation.

Table 1Definitions of variables.

Treated as endogenous variablesBOj Total box office receipts from Friday to Thursday for movie j in

the release weekTHEATERj Number of movie theaters engaged for movie j in the release weekSUPERBOWLj Super Bowl TV advertising expenditure by movie jTVAD,j Total regular TV advertising expenditure up to and including the

release week of movie j

Treated as exogenous variablesCOMPj Total production budgets of all movies released in the same week

and one week prior to the release of movie jBUZZ,j Buzz for movie j in its release weeka

LEADj Number of weeks between the first major TV ad and the releaseweek of movie j

PUB,j Cumulative publicity received by movie j, up to and includingthe release weekb

SEQj A binary variable to indicate if movie j is a sequelGENREj 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 jDISTRIBUTORj 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) Othermovie 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 thefollowing 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 amajor U.S. holiday

a The buzz measure is an inverse transformation ofMOVIEmeter from IMDb.com. Basedon which specific movie pages its four to five million weekly visitors view, IMDb.comproduces the weekly MOVIEmeter ranking for more than 290,000 movie titles in itsdatabase. However, IMDb.com does not provideweekly traffic data at its site. So before theinverse transformation, theweekly rankings are adjusted for eachweek'sweb traffic usingestimates from Alexa.com. The detailed derivation of the buzz variable is in Appendix A.

b Wemeasure the publicity for amovie by determining the total amountof coverage themovie received in EntertainmentWeekly, which has a circulation of 1.79 million, the largestafter the numberonepublication, TVGuide, in the entertainmentmagazine category (AuditBureau of Circulations). We first identified articles related to specific movies by coding thetable of contents of each issue of EntertainmentWeekly. We then classified each article intoone of ten categories, e.g., Departments, News & Notes Category I, or Movie Review. Wedetermined the amount of publicity generated by each article using the average number ofpagesof the category towhich it belongs.PUBj is thendefinedas the sumof coverage valuesof all articles for movie j before and including its release week.

c Starting from the twelve genre categories used by Variety.com, we simplified thecategories 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.

6 We also estimated a model with BUDGETj in both equations, but the effect ofBUDGETj 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 themodel with BUDGETj excluded from Eq. (2). Our treatment of budget is similar to thatof EE (2003), and they also did not find budget to be significant for the US first weektheater engagement equation.



to schedule more movies and reduces the operational costs of longerscreening times. On the other hand, anecdotal evidence suggests thatmost consumers are not aware of the length of the movies, and, as aresult, this should not impact movie sales. [3] Distributor/studio ofmovie j (DISTRIBUTORi): Different movie distributors may have dif-ferential strategies and power in dealing with movie theaters. As sucheffects are largely confined to exhibitors, we include the distributorvariables only in the theater equation.

Three sets of variables are unique to Eq. (2), the boxoffice equation:[1] Critics' ratings of movie j (CRITICSj): As critics' ratings are notavailable until just before a movie's release, theater managers cannotuse critics' ratings in making their screening decisions, which need tobe finalized at least a week prior to the movies' releases. [2] Movieseason in which movie j is released (SEASONj): Einav (2007) arguesthat seasonality can only influence demand, but not the number ofavailable theaters for movie screening. This is due to the fact that thenumber of theaters does not change from season to season. [3] Movie jreleased duringone of the eightmajorU.S. holidays (HOLIDAYj): similarto themovie season variable, we expectHOLIDAYj to affect only the boxoffice revenues.

3.2. Estimation challenges

To estimate Eqs. (1) and (2) simultaneously, we use log transfor-mation to linearize the model. We have five endogenous variables onthe right hand side of the equations: TVADj, and SUPERBOWLj in thetheater equation and THEATERj, TVADj, and SUPERBOWLj in the boxoffice equation. The main sources of endogeneity are the potentiallyunobserved (i.e., unobserved by the researchers) characteristics in-fluencing the behaviors of the decisionmakers in amarket: [1] amoviedistributor/studio with a movie of certain unobserved characteristicsis more likely to adopt Super Bowl advertising and/or commit to acertain level of regular advertising; [2] theater managers are morelikely to screen a movie having these unobserved characteristics, and[3] moviegoers are more likely to watch a movie having these un-observed characteristics in the opening week. An example of suchunobserved movie characteristics is the level of special effects in amovie. As this characteristic is unobserved to us and thereby absentfrom our model, there is a potential bias in the estimated effects of theendogenous variables.7

We address the potential endogeneity of our focal variables in twoways. First, we include a comprehensive set of observable moviecharacteristics in our model (see the above discussion of X's and Z's).These observable movie characteristics are usually the cues theatermanagers and consumers use to infer a new movie's appeal. Second,we control for any other unobserved effects of movie characteristicson our focal variables by using a two-stage instrumental variableestimation process. We discuss the details of our estimation procedurein Section 5.

4. Data description

During the period of our study (i.e., 2000–2002) 1445movies werereleased. The majority of these movies received limited releases,implying that they were shown in only a few local markets in the firstweek. None of the Super Bowl movies during this period had a limitedrelease; all received a wide release across US. This is expected giventhe magnitude of the Super Bowl expenditure and its broad reach.Consequently, given that only movies with a major national releaseare likely to advertise during the Super Bowl, we limit our analysis towide-release movies. Following Einav (2007), we chose moviesreleased in at least 600 theaters, leaving 402 movies in the sample.

Even within these 402 movies, there is still great variation in thenumber of theaters in which the movies were screened.

The second sampling criterion is the film's production budget. Asmovies with low production budgets are unlikely to be able to investin Super Bowl advertising, we dropped them from our analysis. Inparticular, our sample consists of movies with a production budget of$15 million or more.8 We also dropped four movies due to missingdata on production budgets, resulting in a sample of 302 movies. Forthese 302 movies, we still observe substantial variation in productionbudgets. In sum, our sample accounts for 79% of the total NorthAmerican box office revenues ($25 billion) for all movies releasedfrom 2000 to 2002.9

We identified the TV commercials for individual movies placed inthe Super Bowls of 2000, 2001 and 2002 from TV recordings of theSuper Bowl games from kickoff to the end of the game. While therewere TV commercials for other films appearing before and after thegames (e.g., during the pre-game show), we included only thecommercials appearing in the commercial breaks during the games.This is the standard definition of Super Bowl advertising in theacademic and trade literature. There were 19 movies advertisedduring the Super Bowls from 2000–2002.10 While all of the SuperBowl-advertised movies placed a single 30-second commercial duringthe Super Bowl (except Mission to Mars, which aired a 60-secondadvertisement), there is clear variation in such other variables asproduction budget and launch TV advertising spending.

As shown in Table 1, the variables in our study are constructed fromseveral 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 morethan 25 million visitors each month; [3] TNS/CMR, the researchcompany tracking TV commercials on over 425 network and cablechannels in more than 75 TV markets in the United States; [4] Enter-tainment Weekly, the popular consumer magazine for entertainment;[5] Rottentomatoes.com, a comprehensive website archiving reviewsby movie critics; and [6] Alexa, an Amazon.com subsidiary that tracksweb surfing of Internet users based on their proprietary search tool

7 For further discussion on omitted variables and endogeneity, please refer to Greene(2003, page 334).

8 In our sample of movies that advertised during the Super Bowl, the minimumnumber of theaters in which any movie was released was 2225 and the minimumproduction 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 moviesreviewed by selected critics in Variety.10 Details of the Super Bowl movies are available from the authors upon request.

Table 2Summary statistics of Non-Super Bowl (NSB) vs. Super Bowl (SB) advertised movies.

Variables Group Mean Std.Deviation

Minimum Maximum

Box office revenues($ in millions)

SB Movies 42.81 29.32 10.54 110.56NSB Movies 23.89 23.09 1.80 151.62

Number of theaters(in thousands)


Regular TV advertisingspending ($ in millions)


Production budget($ in millions)


Advertising lead time(in weeks)

SB Movies 15.84 9.89 3 39NSB Movies 5.48 4.63 0 34

Movie runtime(in Minutes)

SB Movies 114.00 16.65 94 155NSB Movies 107.29 18.35 72 183

Competition intensity($ in millions)


Pre-launch buzz SB Movies 4.45 1.01 0.37 5.12NSB Movies 4.06 1.01 0.01 5.26

Publicity SB Movies 4.74 5.58 0.00 21.70NSB Movies 2.87 4.06 0.00 31.55

Critics' rating SB Movies 5.25 0.94 3.90 7.00NSB Movies 5.14 1.31 2.20 8.50



Table 3Correlation matrix.

Variables 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31

1 ln(Number of Theaters) 1.002 ln(Box Office Revenues) 0.63 1.003 ln(SB Advertising

Spending)0.29 0.20 1.00

4 ln(Regular TV AdvertisingSpending)

0.78 0.60 0.09 1.00

5 ln(Competition Intensity) 0.09 0.08 −0.06 0.19 1.006 ln(Production Budget) 0.72 0.52 0.18 0.58 0.02 1.007 ln(Movie Runtime) 0.08 0.28 0.10 0.26 −0.01 0.38 1.008 ln(Advertising Lead Time) 0.48 0.40 0.35 0.38 0.08 0.34 −0.04 1.009 ln(Pre-Launch Buzz) 0.26 0.16 0.04 0.25 −0.06 0.17 0.05 0.06 1.0010 ln(Publicity) 0.26 0.43 0.09 0.35 0.00 0.37 0.31 0.15 0.06 1.0011 Binary: GENRE—Action 0.24 0.15 0.19 0.10 −0.11 0.22 0.07 0.18 0.06 0.01 1.0012 Binary: GENRE—Comedy −0.22 −0.19 −0.10 −0.07 0.06 −0.29 −0.31 −0.08 −0.07 −0.10 −0.42 1.0013 Binary: GENRE—Drama −0.19 −0.05 −0.06 −0.06 −0.03 −0.01 0.43 −0.23 0.05 0.10 −0.40 −0.48 1.0014 Binary: MPAA-PG Rated 0.16 0.10 −0.01 0.08 0.00 0.11 −0.18 0.16 −0.02 0.04 −0.12 0.04 −0.21 1.0015 Binary: MPAA-PG13 Rated 0.09 0.04 −0.01 0.11 0.03 0.04 0.05 0.01 −0.02 −0.04 0.14 0.12 −0.07 −0.37 1.0016 Binary: MPAA-R Rated −0.26 −0.12 0.05 −0.13 −0.10 −0.10 0.18 −0.14 0.08 0.07 0.01 −0.15 0.28 −0.26 −0.70 1.0017 Binary: Distributor—Disney −0.09 0.03 0.03 −0.03 0.05 −0.04 −0.12 −0.02 −0.03 −0.09 −0.11 0.07 −0.05 0.08 −0.05 −0.13 1.0018 Binary: Distributor—AOL 0.05 −0.04 0.01 −0.08 −0.02 0.02 0.10 −0.01 −0.07 0.00 0.06 −0.07 0.05 −0.02 −0.12 0.19 −0.26 1.0019 Binary: Distributor—

Viacom0.02 0.00 −0.09 0.11 0.02 −0.01 −0.01 −0.09 0.05 0.01 0.01 −0.07 0.06 −0.03 0.01 0.00 −0.16 −0.20 1.00

20 Binary: Distributor—Sony 0.10 −0.02 −0.03 0.08 0.01 −0.01 −0.04 0.08 0.05 −0.10 0.07 0.02 −0.05 −0.09 0.13 −0.04 −0.18 −0.23 −0.14 1.0021 Binary: Distributor—Fox 0.05 0.03 −0.05 0.02 −0.01 0.00 0.01 0.04 0.02 0.03 −0.03 0.03 0.00 0.07 0.02 −0.03 −0.16 −0.20 −0.12 −0.14 1.0022 Binary: Distributor—

Vivendi0.02 0.13 0.11 0.05 −0.01 0.09 0.01 0.12 0.07 0.13 0.03 0.02 −0.01 0.03 0.03 −0.02 −0.14 −0.18 −0.11 −0.13 −0.11 1.00

23 Binary: Distributor—DreamWorks

−0.12 0.03 −0.01 0.07 0.03 0.05 0.02 0.04 −0.09 0.09 −0.03 −0.04 0.00 0.03 −0.01 −0.04 −0.12 −0.15 −0.09 −0.10 −0.09 −0.08 1.00

24 Binary: Star Power 0.22 0.22 −0.04 0.22 0.09 0.20 0.23 −0.02 0.07 0.16 −0.12 −0.02 0.15 −0.02 0.00 0.03 −0.06 −0.06 0.10 −0.04 0.03 0.01 0.14 1.0025 Binary: Director Power 0.20 0.21 −0.04 0.15 0.06 0.19 0.22 0.05 0.05 0.11 0.00 −0.07 0.06 0.07 0.08 −0.12 −0.08 0.05 −0.06 −0.07 0.14 0.02 0.04 0.29 1.0026 Binary: Sequel 0.39 0.33 0.15 0.09 0.03 0.20 −0.03 0.17 −0.01 0.08 0.12 0.01 −0.20 0.12 −0.10 −0.06 0.06 −0.01 0.03 −0.07 −0.04 0.09 −0.08 −0.07 0.09 1.0027 ln(Critics Rating) 0.25 0.46 0.04 0.37 0.16 0.25 0.32 0.18 0.05 0.34 −0.11 −0.15 0.10 0.12 −0.11 0.00 0.08 −0.12 0.01 −0.13 0.05 0.08 0.07 0.21 0.22 0.08 1.0028 Binary: SEASON—Jan–Apr −0.21 −0.23 0.10 −0.24 −0.31 −0.20 −0.08 −0.17 −0.03 0.06 −0.03 0.08 0.00 0.01 −0.12 0.15 0.03 −0.02 0.03 −0.05 −0.06 −0.02 −0.08 −0.07 −0.11 −0.06 −0.10 1.0029 Binary: SEASON—May–Aug 0.13 0.26 0.09 0.11 0.22 0.14 −0.02 0.28 −0.06 0.00 0.03 0.02 −0.10 0.06 0.08 −0.13 0.00 −0.05 −0.06 0.06 −0.01 0.03 0.12 −0.02 0.00 0.11 0.04 −0.49 1.0030 Binary: SEASON—Nov–Dec 0.23 0.11 –0.13 0.24 0.29 0.20 0.14 0.01 0.03 0.01 0.02 –0.04 –0.04 –0.01 0.06 –0.12 0.05 0.02 0.04 0.01 0.04 −0.04 −0.06 0.16 0.17 0.16 0.12 −0.33 −0.38 1.0031 Binary: Holidays 0.09 0.15 −0.08 0.09 0.15 0.12 0.19 −0.01 −0.01 0.01 −0.05 −0.06 0.07 0.03 0.03 −0.10 0.06 0.04 −0.01 0.00 −0.09 −0.04 −0.04 0.09 0.15 0.01 0.06 −0.05 −0.09 0.32 1.00

⁎ Highlighted numbers are significant at 5% level.

173J.Y.C.H

oet

al./Intern.J.of

Researchin

Marketing

26(2009)

168–179


bars. Our operationalization of several variables departs from theliterature (e.g., BDT 2006). For example, we measure pre-launch buzzas the amount of interest in the movie on the IMDb website, ascompared to using the number of theaters showing the movie orrevenues per screen as an indicator of post-consumption word-of-mouth. We also introduced several new control variables, such as thelead time from the start of the movie's launch TV advertisingcampaign to its release date (LEADj) and the pre-launch publicityreceived (PUBj), as measured by coverage of the film in EntertainmentWeekly.

In Table 2, we present summary statistics for the continuousvariables for the Super Bowl and non-Super Bowl advertised movies.As expected, Super Bowl movies have higher average box officerevenues and theater engagement numbers in the opening week ascompared to non-Super Bowl movies, even though our sampleconsists of only wide-release movies. Super Bowl movies also havelarger launch TV advertising expenditures (TVADj), higher productionbudgets (BUDGETj), a longer lead time from the first major TVadvertising effort to release week (LEADj) (as in our earlier discussionof Fig. 1), longer runtime (RUNTIMEj), less competition (COMPj), ahigher level of pre-launch buzz (BUZZj), more publicity (PUBj), andbetter critics' ratings (CRITICSj), suggesting that these variables arepotential confounding factors with the use of Super Bowl TV ad-vertising. In terms of rating, Super Bowl movies tend to be PG13 or R-rated action movies released between February and August. Com-pared to non-Super Bowl movies, Super Bowl movies are more likelyto be sequels.11 Table 3 presents the correlations among the keyvariables in the study. None of the correlation estimates among ourcontrol variables are high in terms of magnitude. To test whethermulticollinearity is an important factor in our results, we calculatedthe Variance Inflation Factor (VIF) for the variables in Eqs. (1) and (2).A standard rule of thumb (Belsley, Kuh, &Welsch, 1980) is that the VIFbe 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 thecomplex nature of the endogenous variables in our model. Specifically,Super Bowl advertising decisions can potentially be endogenous. Toovercome this problem,we first estimate a probit model to approximatethe Super Bowl advertising decision using a set of exogenous variables(Stage 1). Bymultiplying the probit's predicted probability of advertisingin the Super Bowl by the unit cost of Super Bowl advertising expenditure(i.e., the cost of a 30-second advertisement), we obtain the expectedSuper Bowl expenditure for each movie. We then substitute thisestimated expenditure for SUPERBOWLj when estimating Eqs. (1) and(2) (Stage 2). These predicted or expected values only contain the part ofthe variation in SUPERBOWLj that is due to the exogenous variables,which are uncorrelated with the errors in Eqs. (1) and (2); thus, thepotential endogeneity due to unobserved characteristics of SUPERBOWLjis removed from the model.12 As shown in Mroz (1999), two-stageprocedures such as ours will result in consistent and asymptoticallyunbiased estimates. To validate our results, we employ several tests tocheck for the consistency and appropriateness of our instruments toestimate themodel. UnlikeBDT (2006),wedonot specifyanequation fortotal regular launch TV advertising expenditure. Instead, we control forthe endogeneity of regular TV advertising TVADj by the use of the

instrumental variable method. We abstract away from specifyingadditional equations for the movie distributors'/studios' advertisingdecision processes in order to keep the model parsimonious and trac-table, and also to focus on the two critical outcomes of theaterengagements and box office revenues. This approach also helps reduceunintentional bias from directly estimating the advertising decisionprocess. A similar approach has been used in empirical demand analysisto avoid obtaining biased parameter estimates in new empiricalindustrial organization studies (Dube & Chintagunta, 2003).

The variables used in the probit model are BUDGETj, GENREj,MPAAj, DISTRIBUTORj, SEQj, HOLIDAYj and the time difference in weeksbetween the Super Bowl week and theweek preceding the release of amovie (denoted as SB_DISTANCEj):

ProbðAdvertising in Super Bowl by movie jÞ= Φðα0 + α1⋅BUDGETj + α2⋅BUDGET

2j + α3⋅SB DISTANCEj

+ α4⋅HOLIDAYj + α5⋅SEQj + ∑kαk⋅GENREjk + ∑

mαm⋅MPAAjm

+ ∑nαn⋅DISTRIBUTORjnÞ

ð3Þ

where Φ(.) is the cumulative distribution function of the standardnormal distribution.13

Super Bowl movies are neither the most expensive nor the leastexpensive of those released by the distributors/studios. This impliesthat there can be nonlinear effects of the production budget on SuperBowl advertising decisions. To control for such a non-linear effect, weadded squared BUDGETj as an explanatory variable. In addition, as noSuper Bowl movies in our sample were G-rated or distributed byViacom, we dropped the corresponding dummy variables in our es-timation. To avoid perfect multicollinearity, we also dropped onebinary of each set of the categorical variables (specifically, theindicators of family movies and movies distributed by smalldistributors).

Table 4 presents the probit regression results. Of the variables usedto specify the probit model, BUDGETj, BUDGETj2, SB_DISTANCEj, and thedistributor binaries for Disney and Vivendi are significant at pb0.05.The significance of the budget variables suggests that there is a non-linear effect of the production budget on the Super Bowl advertisingdecision. Up to a level of $49 million, the higher the production bud-get, the more likely a movie is to be advertised during the Super Bowl,but after that the likelihood declines. As expected, the negative andsignificant SB_DISTANCE suggests that the longer the time betweenthe Super Bowl and the movie release date, the less likely is SuperBowl advertising. The significance of Disney and Vivendi suggests thatdistributors can have specific and significantly different strategies interms of Super Bowl advertising.

There are three other endogenous variables on the right hand sideof the model (i.e., THEATERj in Eq. (2) and TVADj in both equations).We control for their endogeneity by two instrumental variables inaddition to the exogenous variables already included in Eqs. (1) and(2). They are [1] BUZZt−1, j: the buzz in the week before the releaseweek, and [2] TVADt−1, j: the cumulative regular advertising expen-diture in the week before the release week. Specifically, in the theaterengagement Eq. (1), to control for the endogeneity of TVADj, we addboth TVADt−1,j and BUZZt−1, j (i.e., ln(TVADt−1, j) and ln(BUZZt−1, j) inthe linearized Eqs. (1) and (2)). In the box office revenue Eq. (2), tocontrol for the endogeneity of THEATERj and TVADj, besides usingTVADt−1, j and BUZZt−1, j, we also use the instrumental variablesRUNTIMEj, BUDGETj and DISTRIBUTORj, which are excluded fromEq. (2) as explanatory variables.

11 Details of Super Bowl and non-Super Bowl movie characteristics are available fromthe authors upon request.12 Heckman (1978) first proposed a similar approach to control for discreteendogenous variables. In labor economics, a similar approach has been widely used(for example: Lee, 1978; Blundell & Powell, 2004; Sandy & Elliot, 1996). In anunpublished working paper in marketing that we were not aware of when weundertook this research, Luan and Sudhir (2007) used a similar approach to control foradvertising endogeneity.

13 As conglomerates own both movie studios and television networks (e.g., Disneyand ABC are jointly owned), we tested to see if movie studios were more likely toadvertise when the Super Bowl was shown on their jointly owned networks, but foundno such effects.



To demonstrate that our approach controls for the endogeneity andprovides consistent estimates, we use a set of diagnostic tests. We firsttest their endogeneity by the Hausman test (Greene, 2003). Similarto BDT (2006), we find that in the second stage estimation, totalregular launch TV advertising and theater engagements are endogen-ous (test statistic=46.72 with p-valueb0.001). We also use a test ofoveridentification to examine the appropriateness of our instruments(Woolridge, 2002) and do not reject the null hypothesis, implying thatour sets of instruments are orthogonal (i.e., not correlated) with theerror structures of the model (test statistic=6.77, p-value=0.75).

In addition to examining the statistical significance of the effect oftheater engagement in the box office Eq. (2), we also demonstrate themediating role of distribution coverage in the Super Bowl advertisingeffect on the opening week box office performance by comparing ourdual path model (DP model) with an alternative model without suchmediation by distribution coverage. Specifically, we drop the theatervariable from the box office equation, creating a non-mediated model(NM model). We use a likelihood ratio test to compare these twomodels under the assumptions of normally distributed errors (Greene,2003, page 409). We reject the null hypothesis that there is nosignificant difference between the twomodels (LR test statistic=4.27,pb0.01). This implies that the unrestricted model (DP) is preferable tothe restricted model (NM). To compare these two models further, weestimated the system R2 (McElroy, 1977). By this measure, the DPmodel fits substantially better than the NM model with a systemR2=0.701 compared to 0.59. These diagnostics suggest that havingthe theater variable playing a mediating role in the DP modelsignificantly increases the explanatory power of the DP model overthe NM model.

6. Results and discussion

Columns A & B in Table 5 present the parameter estimates of the DPand NMmodels. 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 restrictedversion of the procedure we use. Compared to standard 3SLS, theGMM-3SLS approach allows us to specify equation-specific sets ofinstruments, thereby providing flexibility in estimation.14 Greene

(2003) states that in the presence of unknown heteroskedasticity,system of equations estimation using GMM is also more efficient. Forthe purpose of comparison, we also present results for our DP modelusing the standard 3SLS technique in Table 5 (column C). As the tableshows, the parameter estimates for our focal variables, THEATERj,TVADj, and SUPERBOWLj, are qualitatively similar across the twodifferent estimation methods. In terms of system R2, the GMM-3SLSmodel fits slightly better than the 3SLSmodel. Given the flexibility andefficiency of the GMM-3SLS approach, we use the GMM 3SLS resultsfor the rest of the analysis.

One binary variable of each set of categorical variables, GENRE,MPAA, DISTRIBUTOR and SEASON, is dropped to avoid perfect multi-collinearity in the estimation process. In particular, we drop the binaryvariables that indicate: [1] the movie is a family movie, [2] it is rated Gby the MPAA, [3] it is distributed by one of the smaller distributors,and [4] it is released in the September–October season. When in-terpreting the effects of binary variables, such aswhether amovie is R-rated, we should note that the estimated parameter associated withMPAA-R captures the effects of MPAA-R relative to the base case, aG-rated family movie released in September–October (by a smalldistributor if in the theater equation). The highlighted parameterestimates in Table 5 are significant at the 5% level, and t-test statisticsare generated using robust standard errors. For comparison purposeswe also estimated a model without any control for endogeneity.15 Thevalues of the parameter estimates and significance levels change quitedramatically once we control for endogeneity; more parameters be-come significant once we control for endogeneity, suggesting animprovement in statistical efficiency (i.e., a decrease in estimatedstandard errors). This is not unusual as Villas-Boas and Winer (1999)showed that controlling for endogeneity improves the efficiency ofparameter estimates.

6.1. Hypothesis testing

InH1a,wehypothesized that Super Bowl advertising spendingwouldhave a positive direct effect on opening week box office revenue, evenwhen the mediating role of distribution coverage was controlled for. Asβ2,SB is not significantly different from zero at pb0.05, we cannot findsupport for this hypothesis. This result is in contrast to Yelkur et al.'s(2004) exploratory study, which did not control for the endogeneity ofSuper Bowl advertising and the mediating effect of theaters.

In H1b, we hypothesized that movie exhibitors play a mediatingrole in the causal chain from Super Bowl advertising to opening weekbox office revenues. We test this by examining the two parts of theindirect path. The estimate for β1,SB in the theater equation suggests asignificant positive effect of Super Bowl advertising spending onnumber of theaters, establishing the first part of the indirect path. Thesecond part of the indirect path, the estimate for β2,THR in the boxoffice equation, is also positive. Consequently, our estimation resultssupport this hypothesis. Similar to Super Bowl advertising spending,regular TV advertising spending in the pre-launch period influencesinitial box office revenues mainly through the theater engagementfactor. In particular, we found that TVADj's effect on BOj works onlythrough THEATERj (significant positive β1,AD but insignificant β2,AD).

Our second hypothesis, H2, requires us to compare the effective-ness of Super Bowl and regular TV advertising expenditures. Given thelog–log specification, the estimated parameters are also the elasticityestimates. Thus, the elasticity of theater engagements with respect toSuper Bowl expenditures is 0.012 and with respect to regular TVadvertising expenditures is 0.306. Both of these estimates arestatistically significant. On the other hand, in the box office equation,the elasticities of box office revenue with respect to Super Bowladvertising and with respect to regular TV advertising are both14 Woolridge (2002) (pages 196–198) provides a detailed discussion of the key

differences between GMM-3SLS and standard 3SLS techniques in estimating a systemof equations. We also estimated the model using 2SLS, and the estimates are verysimilar to the estimates from 3SLS.

Table 4Parameter estimates of the Probit Model of Super Bowl advertising.

Variables Parameter estimates

Intercept −6.189Budget 0.098Budget2 −0.001Distance between SB & release −0.058Movie runtime 0.015Binary: holidays −0.437Binary: Sequel 0.711Binary: Genre—Action −0.410Binary: Genre—Comedy −0.850Binary: Genre—Drama −1.157Binary: Distributor—Disney 1.312Binary: Distributor—AOL/TW 0.635Binary: Distributor—Sony 0.876Binary: Distributor—Fox 0.432Binary: Distributor—Vivendi 1.020Binary: Distributor—Dream Works 0.308Binary: MPAA-PG13 0.699Binary: MPAA-R 0.938Pseudo R2 0.653

Highlighted and underlined numbers are significant at 5% level. Model fit measure:McKelvey and Zavoina's pseudo R2=0.653.

15 Detailed results for this model are available from the authors upon request.



insignificant. Nevertheless, the magnitude of the coefficients for TVadvertising is greater than those for Super Bowl advertising. Thissuggests that the first dollar available for advertising should be spenton regular TV advertising. However, the average movie in our samplespends $13 million in TV advertising, while most movies spend zero inSuper Bowl advertising. Given the diminishing marginal returns fromall advertising (all elasticities are smaller than one), initial Super Bowladvertising can provide larger returns for movies than spending at orabove the average expenditure level. To explore this issue further, weconducted a counterfactual simulation analysis.

6.2. Counterfactual simulation

We conducted a counterfactual simulation by giving each movie inour sample an additional $2.2 M, which is equal to the average rate forone 30-second spot during the Super Bowls of 2000–2002, to spendon either Super Bowl or other TV advertising opportunities. Wecreated 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 isused as the benchmark for comparison).

Scenario 1 (with extra Super Bowl): Each movie spends itsadditional $2.2 M on Super Bowl advertising.

Scenario 2 (with extra AD): Each movie spends its additional $2.2 Mon 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 revenuesshows that Scenario 1 provides higher returns than Scenario 2, i.e., themarginal effect of Super Bowl advertising on the first week box officerevenues is larger than the marginal effect of regular TV advertising inmost cases. To show this more clearly, we subdivide our results into thosemovies that did and did not actually advertise during the Super Bowl(Fig. 3b). Spending an additional $2.2 Mon regular TV advertising bynon-SuperBowlmovies increasesfirstweekboxoffice revenues from$23.89 Mto $25.50 M; however, spending the same amount on Super Bowladvertising increases first week box office revenues to $31.37 M. On theother hand, a $2.2 M extra expenditure by a Super Bowl movie on SuperBowl 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 sameamount of money on regular advertising, then the increase in first weekboxoffice revenueswouldhavebeenanaverageof $2.84 Mabove thebasecase. This simulation exercise suggests that Super Bowl movies shouldonly buy one Super Bowl advertisement, which is the usual practice forSuper Bowl movies.

6.3. Additional results of interest

Whenever directional hypotheses are possible, our significant(pb0.05) results are always in the expected direction. We now brieflysummarize 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 GMMestimation. However, as it is significant in the 3SLS estimation, there is a need forfurther research on this interesting variable.

Table 5Parameter estimates of the regression models.

Model (A) (B) (C)

Variables Dual Path Model (DP)-GMM Non-Mediated Model (NM)-GMM Dual Path Model (DP)-3SLS

Theater estimates Box office estimates Theater estimates Box office estimates Theater estimates Box office estimates

Intercept 3.371 5.860 3.375 8.534 3.526 6.253ln(Production Budget) 0.050 0.045 0.040ln(SB Advertising Spending) 0.012 0.008 0.013 0.018 0.015 0.004ln(Regular TV Advertising Spending) 0.306 0.149 0.303 0.457 0.297 0.107ln(Number of Theaters) 0.933 0.968ln(Critics Rating) 0.891 0.748 0.985ln(Competition Intensity) −0.030 −0.164 −0.026 −0.175 −0.038 −0.216ln(Movie Runtime) −0.226 −0.206 −0.175ln(Pre-launch Buzz) 0.016 0.066 0.013 0.073 0.017 0.071ln(Advertising Lead Time) 0.037 0.088 0.038 0.128 0.033 0.080ln(Publicity) 0.000 0.021 −0.001 0.020 −0.001 0.019Binary: Sequel 0.142 0.506 0.142 0.656 0.122 0.510Binary: Genre—Action −0.050 0.244 −0.040 0.142 −0.065 0.261Binary: Genre—Comedy −0.111 0.086 −0.097 −0.076 −0.102 0.096Binary: Genre—Drama −0.059 0.180 −0.054 0.041 −0.076 0.152Binary: MPAA-PG Rated −0.080 −0.065 −0.063 −0.070 −0.071 −0.047Binary: MPAA-PG13 Rated −0.110 −0.018 −0.094 −0.047 −0.113 0.053Binary: MPAA-R Rated −0.168 −0.115 −0.152 −0.189 −0.148 −0.007Binary: Distributor—Disney −0.160 −0.146 −0.164Binary: Distributor—AOL −0.050 −0.050 −0.043Binary: Distributor—Viacom −0.092 −0.081 −0.062Binary: Distributor—Sony −0.070 −0.064 −0.068Binary: Distributor—Fox −0.060 −0.060 −0.061Binary: Distributor—Vivendi −0.136 −0.130 −0.115Binary: Distributor—DreamWorks −0.206 −0.151 −0.200Binary: Star Power 0.075 0.165 0.074 0.241 0.054 0.168Binary: Director Power 0.040 −0.065 0.039 0.008 0.051 −0.090Binary: Season Jan–Apr −0.005 −0.028 0.050Binary: Season—May–Aug 0.406 0.378 0.458Binary: Season–Nov–Dec 0.220 0.147 0.331Binary: Holidays 0.258 0.289 0.234

Over identification test 6.77 (df 10), p-value: 0.75 7.30 (df 11), p-value: 0.77System-R2 0.701 0.59 0.69

Highlighted and underlined numbers are significant at 5% level.



1) Publicity (PUBj): The significant positive effect of publicity on firstweek box office revenues suggests that moviegoers are influencedby the amount of media attention surrounding individual movies.On the other hand, there is no significant effect of publicity ontheaters, suggesting that theater managers are less influenced bypublicity. Although publicity is a new measure in the movieliterature, Hollywood studios spend considerable sums of moneyto generate publicity, and its further study would be valuable.

2) Sequel (SEQj): The sequel binary is positively significant in both thetheater and box office equations (at the 5% level). In other words,sequels appear to have built-in brand equity, making themattractive to movie exhibitors and moviegoers.

3) Lead time (LEADj): The lead time between the major TV campaignstart date and the movie release date has a significant positiveeffect 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 therecent results in Basuroy, Chatterjee, & Ravid (2003).

5) Star Power (S_POWERj): The significant positive effect of starpower on the number of theaters showing the movie in the first

week but not on first week box office suggests that star powermainly influences opening week box office indirectly throughincreasing distribution coverage.

6) Seasons (SEASONj) and Holidays (HOLIDAYj): As expected, moviesreleased in the summer season of May–August, in the holidayseason of November–December, and during a week including amajor holiday fared better in opening week box office revenuethan movies released at other times of the year.

7. General discussion

In this paper we explore the role of major event advertising onmarket outcomes using advertising data from the movie industry.Utilizing a sample of movies from the years 2000–2002, this paperdemonstrates that both Super Bowl and regular TV advertisinginfluence opening week box office revenue, primarily throughincreasing the number of movie exhibitors showing the movie in itsfirst week of wide release. While these results may be specific to themovie market, they suggest that failure to take into account

Fig. 3. a: Simulation results under three scenarios. b: Simulation results under three scenarios by Super Bowl (SB) and non-Super Bowl (NSB) movies.



distribution in predicting the influence of advertising on new productdemand can be problematic.

To the best of our knowledge this is the first paper to differentiatebetween major event and regular TV advertising in econometricallyestimating their comparative effects. In cluttered and fragmentedmedia markets, major event advertising can play a significant rolefor products targeting a national audience and pursuing widedistribution. Our analysis suggests that major event advertising canplay a distinctive role in building distribution and demand.

We find that Super Bowl advertising is not as effective as regularTV advertising when both are evaluated at the same level. In part, thecomparatively lower effectiveness of Super Bowl advertising mayoccur because it is a one-time occurrence that is typically distant fromthe actual showing of the movie; by contrast, regular TV advertisingruns over a number of weeks and typically occurs closer to themovie'sopening. However, Super Bowl advertising is still attractive in manycases, as shown in our counterfactual analysis. It can usefully providebetter returns after a minimum threshold of regular TV advertising isreached. This result suggests that firms with limited budgets may bebetter off using regular television advertising rather than Super Bowladvertising. While our results indicate that the use of Super Bowladvertising for at least some movies with above average advertisingbudgets may be more profitable than increased spending on regularadvertising, we find little support for a movie to invest in more thanone Super Bowl advertisement. In terms of methodology, in this paperwe apply a two-stage instrument variable estimation technique toestimate a model with discrete endogenous explanatory variables. Inthe first step we model the discrete choice process, and in the secondstep we estimate a system of theater and box office revenue equationsafter incorporating the estimated probability of advertising in theSuper Bowl from the first step. This approach helps us to generateconsistent and asymptotically unbiased parameter estimates.

At the same time, we acknowledge several limitations in this paperand believe these limitations open interesting future researchopportunities.

First, it would be interesting to examine the effects of Super Bowladvertising in other product categories. Our results provide insight formarketers seeking to introduce new products and build distribution.For many other categories, Super Bowl advertising is part of anongoing advertising campaign for products already in the market.Thus, it becomes a considerable econometric challenge to separate outthe effects of other advertising and continuing sales in order tomeasure the impact of Super Bowl advertising. On the other hand,such product categories also provide the opportunity to study theeffect of multiple ads for the same brand during one Super Bowl.

Second, our research indicates the critical role of advertising ininfluencing marketing channel intermediaries. To understand theoverall impact of advertising, we need to model the influence ofadvertising on the channel intermediaries and on other stakeholdersin the production and marketing processes. For example, in recentyears, a number of retailers and airlines have featured their ownemployees in their advertising campaigns. Does such advertising firstinfluence productivity (by improving employee morale) and thendirectly or indirectly improves sales?

Many CEOs are questioning the efficacy of large scale advertisingspending, especially in this age of fragmented media markets whereconsumers also control the content towhich they are exposed throughdevices like TiVo. The marketing metrics literature is one response tothis call. Further research that quantifies and provides detailedpathways of effects for both major event advertising and standardadvertising expenditures is, in our view, extremely valuable.

Appendix A. Derivation of buzz variable

We measure pre-launch buzz for a specific movie based on thenumber 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, aweekly measure called MOVIEmeter. MOVIEmeter is a rank measure ofindividual movies' total page views in a specific week that we convert,using an inverse function, to a measure of share of views. To allow acomparison of two equally ranked movies in two different timeperiods (say, one movie ranked second in August 2001 and anothermovie with a second-place MOVIEmeter rank in March 2002), we useadditional assumptions and data sources to convert the rank measureof MOVIEmeter into a traffic-based measure of pre-launch buzz formovie j in its release week t, BUZZjt. Specifically, we use the followingfunction:

BUZZjt =1

MM―Rankjt + C

" #*IMDb―Traffict ðA1Þ

where MM_Rankjt=MOVIEmeter rank for movie j in its release week tand IMDb_TRAFFICt is the estimated number of visits to IMDb.com inmovie j's release week t.

The first part of expression (A1), 1MM―Rankjt + C

h i, is our assumed

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

set C at 60 such that the sum of the share, 1MM―Rankjt + C

h i, over the top

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 couldnot observe the actual number of visits to IMDb.com, we obtainedfrom Alexa.com the daily number of visitors to IMDb.com per onemillion Alexa tool bar users from September 2001 to August 2004.Assuming the proportion of Alexa tool bar users to the total Internetuser population remained the same, we estimated a regression modelcapturing the seasonality of Alexa's IMDb.com traffic and then usedthe regression model to extrapolate backwards the weekly traffic toIMDb.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 con-sumption: An analysis of causality. Econometrica, 48(5), 1149−1168.

Basuroy, S., Chatterjee, S., & Ravid, S. A. (2003). How critical are critical reviews? Thebox 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 inthe motion picture industry. Journal of Marketing Research, 43, 297—295.

Belsley, D. A., Kuh, E., & Welsch, R. E. (1980). Regression diagnostics: Identifying influentialdata and sources of collinearity. New York: Wiley.

Blundell, R. W., & Powell, J. L. (2004). Endogeneity in semi parametric binary responsemodels. Review of Economic Studies, 71(3), 655−679.

Chu, W. (1992). Demand signaling and screening in channels of distribution. MarketingScience, 11(4), 327−347.

Desai, P. S. (2000). Multiple messages to retain retailers: Signaling new productdemand. 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 releasedproducts 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). Dummyendogenous 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). Unionismandwage rates: Simultaneous equationsmodelwith 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 self-selecting members. International Journal of Research in Marketing, 24, 31−47.



Luan, Y. J. & Sudhir, K. (2007). Forecasting Advertising Responsiveness for Short Life-Cycle 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 importanceof advertisement frequency, length and placement on recall. Journal of MarketingCommunications, 4(4), 237−248.

Newell, S. J., Henderson, K. V., & Wu, B. T. (2001). The effects of pleasure and arousal onrecall 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 investigationinto the relationship among program context, emotional experience, and ad recall.Journal of Consumer Research, 15(3), 360−367.

Russell, C. A. (2002). Investigating the effectiveness of product placements in televisionshows: The role of modality and plot connection congruence on brandmemory andattitude. Journal of Consumer Research, 29(3), 306−318.

Sandy, R., & Elliot, R. F. (1996). Unions and risk: Their impact on the level of compensationfor fatal risk. Economica, 63(250), 291−309.

Tomkovick, C., Yelkur, R., & Christians, L. (2001). The USA's biggest marketing eventkeeps 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. ManagementScience, 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.


Author's personal copy - UBC Sauder School of Business · thousand viewers (CPM) of $19.85,...

Documents

Transcript of Author's personal copy - UBC Sauder School of Business · thousand viewers (CPM) of $19.85,...