Super Bowl Ads∗

Wesley R. Hartmann Daniel KlapperGraduate School of Business School of Business & Economics

Stanford University Humboldt University Berlin

March, 2015First Draft: June, 2012

Abstract

We explore the effects of television advertising in the setting of the NFL’s Super Bowl telecast.The Super Bowl is the largest advertising event of the year and is well suited for measurement. Theevent has the potential to create significant increases in “brand capital” because ratings averageover 40 percent of households and ads are a focal point of the broadcast. Furthermore, variationin exposures is exogenous because a brand cannot choose how many impressions it receives in eachmarket. Viewership is determined based on local preferences for watching the two competing teams.With this significant and exogenous variation in Super Bowl advertising exposures we test whetheradvertisers’ sales are affected accordingly. We run our analysis using Nielsen ratings and store levelsales data in the beer and soda categories. We find that Super Bowl ads can generate significantincreases in revenue per household. However, when two major soda brands both advertise, muchof this gain is lost. Exploring the mechanism behind the ad effects, we find that Super Bowl adsbuild an association between the brand and viewership of sports more broadly.

Keywords: advertising, branding, complements, television.

∗Email addresses are wesleyr@stanford.edu and daniel.klapper@hu-berlin.de. The authors would like to thank LanierBenkard, Latika Chaudhary, Jean-Pierre Dube, Matt Gentzkow, Ron Goettler, Brett Gordon, Peter Rossi, Jesse Shapiro,Ken Wilbur, seminar participants at UC Davis and Michigan, conference participants at the 2012 Marketing ScienceConference, the 5th Workshop on the Economics of Advertising and Marketing in Beijing, the 2013 Summer Institute inCompetitive Strategy and 2014 NBER IO Summer Institute. Corey Anderson and Chris Lion provided valuable researchassistance on this project.

1

1 Introduction

The Super Bowl is the premier advertising event of the year. Four of the five most watched telecasts

ever were Super Bowls. The 2012 broadcast was the most watched telecast in history at 54% of US

households. Costs of airing a thirty second spot during the game have grown to more than $3 million.

The two biggest spenders have been Anheuser-Busch (Budweiser) and Pepsi1: both well-known brands

whose existence and tastes presumably do not need to be communicated. This highlights one of the

most puzzling questions in advertising. Can continued heavy advertising by established brands pay

off, and if so why?

The Super Bowl presents an interesting laboratory to answer these questions because it resolves

two of the most significant measurement challenges. First, most television advertising studies lack

credible sources of exogenous variation to infer causality. This is particularly problematic in advertising

because the growth of data informed marketing has made it easier for media planners to endogenously

concentrate ad exposures to their best potential targets. Media planners cannot however control the

number of Super Bowl ad exposures they receive in a given market. It is driven by preferences to

watch the two competing teams. Advertisers can choose whether to buy an ad or not, but most spots

are sold before the season even begins.2 Second, advertising effects may be quite small and difficult to

detect, especially in the context of well-known brands with a long history of advertising and consumer

experiences. Advertising opportunities that can create a substantial shift in this “brand capital” are

rare. But, if there is a single advertising event large enough to shift preferences for established brands,

the Super Bowl is it. Consider the following example. When the Green Bay Packers returned to the

Super Bowl after 13 years in 2011, an additional 14 percent of households in Milwaukee watched the

game. That exposed more Wisconsinites and Packers fans elsewhere to perennial advertisers. If those

ads are effective, we should see perennial Super Bowl advertisers exhibit a corresponding increase in

their sales.

We construct a panel data set consisting of nearly 200 media markets and 6 years of Super Bowl

ratings and sales data from Nielsen. We find that advertising alone in the Super Bowl increases revenue1http://sports.espn.go.com/nfl/news/story?id=47514152Competition for spots in 2010 led 80% of capacity to be sold out eight months in advance (AdWeek, 2012)

2

per household by 10 to 15 percent in the eight weeks following. This implies a return on investment

(ROI) between 150 to 170 percent,3 but these gains turn to losses if a competitor also advertises

during the same Super Bowl. When we observe both Coke and Pepsi advertising, the lift in revenue

per household from advertising vs. not is cut by half or more. Considering that non-advertisers may

incur a decrease in revenue per household, we find evidence suggestive of a prisoner’s dilemma in which

both soda brands advertise, but realize a negative return on their advertising investment.

With large potential returns, it is useful to explore how advertising influences sales for established

brands. Major advertisers such as automobile companies may communicate new products and pro-

motional offerings in their ad spots, but consumables such as these beverage brands, other consumer

packaged goods companies, and even quick serve restaurants are well-known with relatively stable prod-

uct lines.4 We find evidence that the ads work by creating or enhancing a complementarity between the

brand and potential consumption occasions. Specifically, the greatest returns to Super Bowl ads occur

in weeks when shoppers make purchases to consume during other major sports broadcasts. To test

whether a Super Bowl ad yields a complementarity between the brand and the viewership of sports,

we collected market week level data on viewership of the NCAA basketball tournament and interacted

it with the Super Bowl ad exposures. We find this to be a significant determinant of the effectiveness

of Super Bowl ads. This mechanism is consistent with the lack of returns when both advertise head

to head in the Super Bowl, as it may be difficult for two major brands to simultaneously capture the

same association.

We are not the first to consider a role for advertising in forming complements. Becker and Murphy

(1993) propose advertisements themselves complement consumption. Our finding is that the advertising

builds complementarity between the brand and something else such as the way potential consumers

are spending their time. The volume of sports viewership in our society clearly suggests itself as

a valuable complement. But other brands may differentiate by choosing other complements. For

example, Corona beer associates itself with relaxation and romance through imagery of a couple on a3Extrapolations to ROIs are highly approximate because the actual prices paid for the ads are unknown and the

scaling of sales data to the appropriate level are subject to numerous assumptions. We detail this in the results sectionof the paper.

4Some of the Super Bowl ads in our analysis focus on smaller sub-brands, such as when Diet Pepsi Max had recentlybeen rebranded to Pepsi Max, but the majority emphasize just the brand, or well established product offerings such asBud Light (the largest selling beer).

3

beach. The same person may want a Bud when watching a game with friends, but a Corona when with

a (potential) partner. In fact, the most recent ads for Budweiser emphasize “Grabbing some Buds”

within many social contexts, perhaps competing to be the beer of choice whenever a guy is with his

buddies. The key differentiation from Becker and Murphy (1993) being that brand sales are influenced

by changes in demand for the complementary activity even when advertising is held fixed. Advertising

complementarity may therefore arise indirectly through the associations it creates with the brand.

An interesting feature of this complementarity is that we find it in the soda category where the

“creative” of the advertising rarely emphasizes sports. This suggests that the context in which the ad

airs can play an important role in generating the complementarity. This has implications for market

structure in the advertising industry. The recent deconcentration of consumer viewing habits both

within television and to the abundant online alternatives threatens to commoditize the advertising

market. However, if the context in which the ad is viewed is an important part of its effectiveness, as

we find here, then television channels, publishers and websites can differentiate themselves within this

market based on the content they produce or acquire and distribute.

It is useful to consider our findings in the context of other studies of advertising and the Super Bowl.

The challenges of measuring advertising effects is nicely described in Lewis and Rao (2014). Considering

field experiments for internet advertising, their primary point is that effective advertising can involve

very small changes in sales. But, detection of small effects requires a very large sample size if there is

considerable variance in sales. They point out that this same “power” issued led TV ad effectiveness

studies to report findings at the 80 percent confidence level (e.g. Lodish et.al. 1995). These challenges

have been overcome by recent experimental studies on direct mail advertising (Bertrand et.al., 2010)

and internet advertising (Sahni, 2013), but there is still a dearth of studies analyzing TV advertising

with credible sources of exogenous variation. Our Super Bowl analysis overcomes these concerns about

statistical power because of the large potential effects and the market level data which includes millions

of households and shifts inference away from the high variability of sales that might be observed in a

cross-section of household-level sales. Furthermore, market level fixed effects shift inference to within

markets where Bronnenberg et.al. (2009) have documented less variation in brand performance.

There are also a couple of papers that explore Super Bowl advertising specifically. Lewis and

4

Reiley (2013) studies the effect of Super Bowl ads on search behavior and finds a significant spike

within seconds of the airing. Such immediate search effects do not necessarily imply sales effects and

would include viewers’ desires to either see the commercial again or follow up on something from the

ad. The only other paper tying exogenous variation in Super Bowl ad exposures to sales is Stephens-

Davidowitz et.al. (2013). They apply a modified version of our identification approach to the case of

movies.5 They find significant positive effects for movies released well after the Super Bowl. Effects

in movies likely represent a strong informative6 or free-sample component, whereas our focus is on

the effects and mechanism of advertising by familiar brands with established advertising stocks (as

in Nerlove and Arrow, 1962, Dube, Hitsch and Manchanda, 2005, Doganoglu and Klapper, 2006, or

Doraszelski and Markovich, 2007).

The remainder of the paper proceeds as follows. The next section describes the data sources.

Section 3 presents the estimates and section 4 concludes.

2 Data

We analyze the relationship between within market variation in Super Bowl ratings and within market

variation in Super Bowl advertisers’ sales. The ratings data for the top 56 Designated Market Areas

(DMAs) is publicly released in some years,7 but was purchased from Nielsen. We also obtained access

to the AdViews database to collect SB ratings for the remainder of the DMAs allowing us to include up

to 195 markets. AdViews also provides weekly advertising exposures for brands as well as market-level

exposures to ads during NFL broadcasts leading up to the Super Bowl.

Data on store level revenue and volume, as well as trade data (feature and display), come from the

Kilts Center’s Nielsen Retail Scanner Data.8 The timeline for our analysis is the 2006-2011 time frame

for which we observe all of these data sources.5Our abstract describing the value of exogenous variation arising from changes in Super Bowl competitors for iden-

tifying advertising effects was first published in an abstract in the 2012 Marketing Science Conference program. Theauthors can provide a copy of this upon request.

6Ackerberg (2001) and others document informative advertising effects arising for new products.7Nielsen actually releases data for the top 55 DMAs, but because of entry and exit from the top 55, we have 56.8A previous version of this paper was circulated using the IRI Marketing data set described in Bronnenberg, Kruger

and Mela (2008). We switched to the Nielsen data because of an exact match in the definition of the geographic regionand the ability to increase the number of DMAs in our analysis from roughly 33 to 56. The Nielsen data also allows usto include the smaller DMAs where statistical power is stretched.

5

2.1 Super Bowl Advertising and Ratings

We focus on the Super Bowl advertising by beer and soda brands. Anheuser-Busch has spent the most

on Super Bowl advertising and has been the exclusive beer advertiser in the Super Bowl for the entire

time span of our data. Pepsi spends the second most and ran an advertisement in every Super Bowl in

our data except 2007 and 2010. In 2007 they sponsored the half-time show instead.9 The withdrawal

from 2010 was based on a widely publicized refocus of their advertising efforts toward a social media

campaign. Coca-Cola advertised every year from 2007 to 2011, but prior to that had not advertised

since 1998. We later discuss potential concerns about selectivity in the advertising decisions in the

carbonated beverage category.

While there is little to no variation in our data regarding who advertises during a Super Bowl,

Figure 1 depicts substantial variation both across and within the top 56 DMAs in the exposures to

the Super Bowl ads. The bars at the bottom represent the average ratings for each DMA as measured

on the axis to the right. The average is 45.3 percent of a market viewing the Super Bowl, with cross-

market variation from 36.4 to 53.4 percent. The dots above represent the year by year deviations

from the DMA mean ratings as measured on the axis to the left. The DMAs are ordered left to right

with increasing variance in the ratings such that the DMA with the smallest variation across years

sees movement of roughly plus or minus 2.5 percent around its average ratings of 47.7 percent. Many

DMAs experience ratings dispersion of ten points or more, while the most variable DMAs experience

ratings dispersion of 17 points. Overall, this paints a picture of large swings in terms of how many

people are watching the Super Bowl and getting exposed to the ads.9We’ve run our analysis also coding this up as an advertisement and while the coefficients change some, the substantive

conclusion remains the same.

6

30.0

40.0

50.0

60.0

70.0

80.0

90.0

100.0

-30.0

-25.0

-20.0

-15.0

-10.0

-5.0

0.0

5.0

10.0

15.0

DMA

Aver

age

Ratin

gs

Ratin

gs D

iffer

ence

from

DM

A Av

erag

e

Figure 1: Super Bowl Ratings by DMA: Average and Year-Specific Deviations

From the advertisers’ perspective, this variation in exposure is out of their control. It is based on

local variation in the preferences for watching the Super Bowl. While it is impossible to decompose all

of the factors generating viewership swings for the Super Bowl, we have tried to identify some of the

most significant elements. Figure 2 illustrates the role of local preferences for the teams in the Super

Bowl in generating ratings. The horizontal axis represents the percentage of the people in the DMA

who “liked” the two teams in the Super Bowl on Facebook as of April 2013.10 The vertical axis plots

the associated ratings for the Super Bowl. The relationship is clearly non-linear with more than 45

percent of the population watching whenever at least 5 percent of the market likes the teams. Among10While a better measure would include how many local people liked the teams before the respective Super Bowl, we

are unaware of a historical source of such information that can date back to 2006.

7

those observations with less than 5 percent liking the team, there is still a correlation of 0.33 with the

observed ratings. This illustrates that preferences for the teams playing is still a significant driver of

viewership even outside the home cities.

This certainly does not explain all of the variation in ratings, as Facebook is not demographically

representative and we should ideally have these preference measures at the beginning of our data. To

quantify the explanatory power, we ran a simple regression of log ratings on the log of the percent

of the DMA liking the team. The R-squared is 0.24. Including both DMA and year fixed effects,

we find that the percent liking the team explains 12 percent of the within-DMA variation in ratings.

Year fixed effects alone explain about 42 percent of the within-market variation. The remainder of the

variation may arise from other unobserved components of the local preferences for the game. Some of

that variation occurs because we often measure likability of the teams well after the game was actually

played.

8

30

35

40

45

50

55

60

0 5 10 15 20 25 30 35

Perc

ent o

f DM

A th

at W

atch

ed th

e Su

per

Bow

l

Percent of DMA that "Likes" Super Bowl teams on Facebook

Figure 2: Super Bowl Ratings by DMA: Average and Year-Specific Deviations

2.2 Retail Scanner Data

Nielsen’s retail scanner data provides unit sales, prices, display and feature information at the UPC

level for each store and week. While the UPCs can be aggregated at many levels (e.g. diet/light vs.

regular, sub-brand or pack size), we report our analysis aggregating sales to the brand level as we

expect that to internalize all of the effects of advertising.

We consider the top four brands (based on volume) in each category while aggregating the remaining

brands in a composite named “Other.”11 In beer, the focal brands are Budweiser, Miller, Coors and

Corona. These four brands also represent the top four purchasers of NFL impressions in weeks prior11To avoid over-weighting large markets in selecting the top brands, we first ranked brands within market and week

and then selected those whose modal ranks were highest.

9

to the Super Bowl. In soda, the focal brands are Coke, Pepsi, Dr. Pepper and Mountain Dew. These

brands also purchased the most NFL impressions prior to the Super Bowl.

We aggregate the store-level data to the geographic level at which we observe ratings, i.e. the

DMA. We consider up to twenty weeks after the Super Bowl, as well as the first five weeks of the year

leading up to the game on the first Sunday in February. The first week after the Super Bowl includes

Super Bowl Sunday. For each week in the data, we have included the DMA level gross rating points

(GRPs) for each brand’s advertising as reported by Nielsen’s AdViews. GRPs /100 are the number of

impressions per capita. We also consider the cumulative GRPs during NFL games leading up to the

Super Bowl. This accounts for brand impressions that would be correlated with Super Bowl ratings

because the Super Bowl competitors also drew large audiences in previous rounds of the playoffs.

Table 1 reports the outcomes and marketing decisions for the twenty weeks following the Super

Bowl. In the beer category, Budweiser clearly dominates their market with a revenue and volume

per household that is comparable to the combination of all brands outside of the top four.12 The

revenue and volume numbers represent only those households that might be covered by the Nielsen

sales data and thus translate into a number smaller than the actual revenue and volume per household.

The data is not necessarily representative as some stores, such as Walmart, choose not to offer their

data to Nielsen. Prices are comparable across the brands with the exception of Corona, an imported

beer priced nearly four cents per ounce more. The beers are featured roughly five percent of the time

(volume weighted), except for the Other category which includes smaller brands that may be unlikely to

promote themselves in stores circulars. Display rates are comparable to feature. Budweiser’s advertises

75% more than its closest competitor Miller with 0.007 impressions per capita. If no household saw

an ad more than once, this would translate to less than one percent of households seeing a Bud ad

in average week. This highlights the importance of the Super Bowl which reaches 45% of households.

Budweiser also purchased the most advertising impressions in NFL games leading up to the Super

Bowl. Coors comes in second at just over half of what Budweiser purchased.

In the soda category, Coke and Pepsi are the market leaders, but the Other category is substantially12The number of households is calculated and held fixed over time based on the median value reported by Nielsen for

the DMA across all six years.

10

larger than either of these brands in terms of both revenue and volume per household. Pricing in soda

is comparable across all brands. Major brands are featured 9 to 10 percent of the time, while the

Other brands are featured about 4 percent of the time. Displays occur 7 to 9 percent of the time for

major brands and just less than 5 percent of the time for Other brands. Pepsi is the leader in terms of

advertising purchased. Pepsi also purchased the most advertising during previous rounds of the NFL

playoffs, with Dr. Pepper following with nearly twice the NFL advertising as Coke.

Table 1: Weekly Summary Statistics: 20 Post-Super Bowl Weeks

BeerVariable Budweiser Miller Coors Corona OtherRevenue per HH 0.322 0.114 0.091 0.057 0.332volume 6 pk 0.077 0.028 0.021 0.008 0.074price per oz 0.059 0.058 0.061 0.098 0.061feature 0.050 0.050 0.053 0.052 0.024display 0.054 0.047 0.049 0.053 0.024GRPs / 100 0.007 0.004 0.002 0.001 0.004Pre-SB NFL GRPs /100 0.164 0.058 0.095 0.009 0.026Observations across 173 DMAs and 20 weeks. Some DMA-years missing.DMAs with alcohold restrictions or negligible sales omitted.

SodaVariable Coke Pepsi Dr Pepper Mtn Dew OtherRevenue per HH 0.381 0.268 0.111 0.144 0.537volume 6 pk 0.209 0.158 0.060 0.077 0.288price per oz 0.026 0.025 0.027 0.028 0.027feature 0.093 0.105 0.021 0.041 0.036display 0.082 0.089 0.069 0.077 0.047GRPs / 100 0.007 0.007 0.003 0.002 0.004Pre-SB NFL GRPs / 100 0.006 0.020 0.012 0.001 0.001Observations across 191 DMAs and 20 weeks. Some DMA-years missing.

3 Empirical Specifications

We now focus on how brands’ weekly outcomes are related to the variation in Super Bowl ratings. We

use the following descriptive regression to analyze the relationship between revenue per household13

(Y ) and Super Bowl advertising (A) and viewership (R).

Yjmyw = α1AjyRmy + α2AjyAkyRmy + δRmy +Xjmywβ + γFE + ξjmt (1)

where j indexes the focal brand, k a competitive brand,m the DMA, y the year and w ∈ {−5, ..., 0, 1, ...}

represents the week relative to the Super Bowl. δ measures how variation in Super Bowl viewership13We have also run the analyses below using volume per household and find the same substantive findings and signifi-

cance patterns.

11

affects all brands. α1 measures any additional impact of viewership for the advertiser. If more than

one brand advertises, α2 measures how competition within the game alters the gains from advertising.

We also include additional weekly brand-market specific covariates, Xjmyw to account for potentially

endogenous responses to the variation in ratings. Most of these are marketing variables summarized

in the previous section. γFE is a set of fixed effects discussed below.

The regression can be estimated separately conditional on each week relative to the Super Bowl

to examine the pattern of effects over time and potential decay. When testing for an average effect,

we pool across weeks. The identification strategy is to consider how within market variation in a

brand’s performance is related to its within market variation in the viewership of its Super Bowl ads.

As illustrated above, within market variation in viewership of the Super Bowl is related to variation

in the preferences for the teams that qualify for the Super Bowl. While it might be possible for such

viewership variation to be correlated with local changes in economic characteristics, and hence category

performance, it is highly implausible that such variation is correlated with the relative preference of

a given brand when it is advertising vs. not, or between advertising and non-advertising brands. To

focus identification on this within market variation, we include fixed effects at the brand-market level.

We also include brand-year fixed effects. This helps ease concerns that, for example, the years when

one of the soda brands might not have advertised is correlated with that being a potentially good or

bad year for sales. Finally, we consider robustness checks analyzing the pre-Super Bowl weeks. If we

combine both pre and post Super Bowl data as defined by w, we include additional fixed effects for

w ≤ 0 and w > 0 and interact these with both the brand-market and brand-year fixed effects. This

assures that the separate estimates for pre and post Super Bowl weeks hold when combining those

weeks into the same specification to test for changes.

3.1 Week-Specific Estimation

We use week specific estimation to evaluate the path of effectiveness over time. The primary variables

of interest for the Super Bowl are the ratings in market m in year y, Rmy, and an indicator Ajy for

whether or not brand j advertised in the Super Bowl in year y. We begin by considering the case of

Budweiser who advertised in every year, i.e. Ajy = 1∀ y . Estimating Equation 1 without covariates, α1

12

measures how Budweiser’s within market performance relates to the within market variation in ratings

as compared to their competitors’. We estimate this separately for each post-Super Bowl week using

brand-market and brand-year fixed effects. In Figure 3, we plot the product of α1 and 0.45 which is the

average ratings for the Super Bowl to obtain Budweiser’s revenue per household increase attributable

to Super Bowl viewership. This is measured relative to competitors’ performance in δ so could reflect

competitors’ losses as well. We separately evaluate this in the average effects and do not find evidence

of a competitive effect of Bud’s Super Bowl advertising. The estimate of α1 is always positive in the

figure. It is statistically significant in a number of weeks with the largest effects occurring in weeks 5,

8, 11, and 17. We test for significance on average in the next subsection, but to gauge the magnitude,

the 0.05 value is 15.5 percent of the average weekly revenue reported in Table 1. Exploring the timing

of the increases and spikes in Super Bowl ad effectiveness, it appears they coincide with subsequent

sporting events. While these movements are subtle, similar but more pronounced spikes in these weeks

appear in the soda category

13

-0.15

-0.10

-0.05

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Reve

nue

per H

ouse

hold

Att

ribut

able

to S

uper

Bow

l Ad

View

ersh

ip

Week After the Super Bowl

Figure 3: Super Bowl Advertising Effect by Week After the Super Bowl: Budweiser

Next, we consider the weekly effectiveness of the Super Bowl ads for the soda category. In soda,

we restrict the analysis to Coke and Pepsi as they are most similar and are both observed in and

out of the Super Bowl. In Figure 4, we similarly plot α1 to evaluate the potential effectiveness over

time. We consider competitive effects in δ and α2 when we evaluate the average effects. The effect is

almost always positive, with significance occurring in weeks 2, 6, 8, 11, 12 and 16. A 0.05 estimate

is 8.5 and 14 percent of the revenue for Coke and Pepsi respectively as reported in Table 1. In soda

there is clearly a large spike in effectiveness in week 8 which corresponds with both the NCAA Final

Four and the opening week of Major League Baseball. The confidence intervals indicate that this is

significantly greater than the advertising effect in the preceding weeks. Other spikes in weeks 6, 11

14

and 16 correspond with the beginning of the NCAA basketball tournament, the NBA Playoffs and

the NBA Finals. We had not hypothesized this relationship, but we test it below in section 3.5 by

including market-week level viewership data for the NCAA tournament.

-0.15

-0.05

0.05

0.15

0.25

0.35

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Reve

nue

per H

ouse

hold

Att

ribut

able

to S

uper

Bow

l Ad

View

ersh

ip

Week After the Super Bowl

Figure 4: Super Bowl Advertising Effect by Week After the Super Bowl: Coke & Pepsi Advertising Alone

We explore the average effects and test for differences in these patterns next. With those, we also

introduce covariates.

3.2 Average Effects Across Post Super Bowl Weeks

To test whether the Super Bowl effects depicted above are significantly positive on average, Table

2 reports estimates of Equation 1 pooling across 8 weeks after the Super Bowl. We begin by sep-

arately considering advertising and non-advertising brands. Then, we combine them in a regression

15

to establish statistical significance of the differential performance of the advertising brands relative

to non-advertisers. The first specification in Table 2 considers only Budweiser. The coefficient indi-

cates that Budweiser receives a 1.09 cent increase in revenue per household for every 10 point increase

in Super Bowl ratings (i.e. ratings are measured as the fraction of the population who watched the

game). The p-value of this estimate is 0.08. Specification (2) repeats the analysis, but includes mar-

keting variables. The effect increases slightly to 1.27 cents for each 10 point increase in ratings and

has a p-value of 0.052. The fit of the regression increases with the inclusion of significant predictors

such as price and feature. Note that we cannot interpret the coefficients on the marketing variables

causally because we do not have exogenous variation in these. Nevertheless, they make clear that

the primary marketing variables of the firm are not exhibiting an endogenous response to the Super

Bowl ad viewership that is driving the estimates. Specifications (3) and (4) replicate the preceding

regressions for all non-Budweiser brands. There is no evidence of a significant increase or decrease in

these brands performance that can be attributed to Super Bowl viewership. Thus, the post-Super Bowl

gains in performance attributable to variation in viewership are exclusive to the advertiser: Budweiser.

Specifications (5) and (6) pool all brands together to illustrate this significant difference for Budweiser

relative to competitors. Specification (5) is precisely the average effect over the first 8 weeks plotted

in Figure 3.

Next, we report the average effects for the soda category in Table 3. The first specification similarly

considers only the major advertising brands: Coke and Pepsi. Super Bowl viewership results in a

statistically significant increase in advertisers’ revenue per household of 0.77 cents for each 10 point

increase in ratings. When both brands advertise in the Super Bowl in the same year, the gain drops

by about half. A non-advertising competitor realizes an insignificant revenue decrease of 0.59 cents for

each 10 point ratings increase. In specification (2), we add covariates and find slightly larger effects

with the same patterns and significance. Specifications (3) and (4) repeat the analysis for brands

other than Coke and Pepsi. In this case, the advertising brand is Dr. Pepper in 2010 who realizes a

significant increase in revenue per household of 0.12 cents for each 10 point increase in ratings. Non-

advertising brands realize insignificant decreases in revenue comparable to Coke or Pepsi when they do

16

Table 2: Effects of Super Bowl Viewership and Advertising in Beer

8 Weeks Post-Super Bowl Included(1) (2) (3) (4) (5) (6)

VARIABLES Bud Only Bud Only Non-Bud Non-Bud All Brands All BrandsPost-Super BowlRatings * Ad 0.109 0.128 0.100* 0.105*

(0.063) (0.065) (0.042) (0.043)Ratings 0.008 0.010 0.008 0.008

(0.035) (0.035) (0.035) (0.036)MarketingGRPs 0.096 0.291** 0.180**

(0.069) (0.045) (0.043)Pre-SB NFL GRPs 0.013 -0.033 -0.004

(0.021) (0.021) (0.017)Price -7.679** -1.342** -2.024**

(1.521) (0.185) (0.307)Price * OtherBrands 0.352 1.055

(0.996) (0.973)Feature 0.202** 0.025** 0.043**

(0.050) (0.007) (0.011)Display -0.029 0.026 0.021

(0.068) (0.015) (0.021)Observations 7,104 7,104 28,411 28,411 35,515 35,515R-squared 0.243 0.32 0.243 0.267 0.243 0.269Number of branddma 173 173 692 692 865 865Fixed effects are included at the brand-market and brand-year.Standard errors are clustered at the market level.** p<0.01, * p<0.05

not advertise. After controlling for covariates, the Dr. Pepper advertising effect is no longer significant.

Specifications (5) and (6) pool all brands together resulting in an advertising effect between that found

for Coke and Pepsi and the smaller Dr. Pepper. After controlling for covariates in (6) just over half

the gain from Super Bowl advertising is lost if a major competitor also advertises.

3.3 Comparison with Pre-Super Bowl Patterns

The above analyses indicate that Super Bowl ratings affect the post-Super Bowl performance of ad-

vertisers relative to non-advertisers. (In soda, the effect also exists within brand when advertising vs.

not). These beverage companies may also exhibit pre-Super Bowl sales relationships with Super Bowl

ratings because consumers buy beverages to consume during the game and the teams competing in

the Super Bowl also played in previous rounds of the NFL playoffs. In the absence of Super Bowl

advertising, these effects should cease with the Super Bowl and termination of the NFL season. In this

section, we test for exactly this. We first document pre-Super Bowl performance relationships with

Super Bowl viewership for both advertisers and non-advertisers. Then we document that pre-Super

Bowl performance effects of non-advertisers die out, whereas advertisers are able to retain or augment

17

Table 3: Effects of Super Bowl Viewership and Advertising in the Soda Category

8 Weeks Post-Super Bowl Included(1) (2) (3) (4) (5) (6)

VARIABLES Coke-Pepsi Coke-Pepsi Not Coke-Pepsi Not Coke-Pepsi All Brands All BrandsPost-Super BowlRatings * Ad 0.077** 0.089** 0.012* 0.007 0.047** 0.044**

(0.013) (0.014) (0.006) (0.006) (0.008) (0.007)Ratings * Ad * Ad -0.036* -0.057** -0.020 -0.026*

(0.015) (0.012) (0.013) (0.011)Ratings -0.059 -0.066 -0.051 -0.042 -0.054 -0.046

(0.037) (0.038) (0.038) (0.039) (0.036) (0.036)MarketingGRPs -0.242** -0.173 -0.294**

(0.063) (0.109) (0.062)Pre-SB NFL GRPs -0.368 0.077 -0.123

(0.214) (0.068) (0.114)Price -15.101** -4.797** -9.063**

(0.907) (0.373) (0.575)Price * OtherBrands 0.633 5.027**

(1.538) (1.535)Feature 0.139** 0.098** 0.168**

(0.033) (0.020) (0.028)Display -0.041 0.022 -0.027

(0.035) (0.015) (0.022)Observations 16,032 16,032 24,048 24,048 40,080 40,080R-squared 0.05 0.305 0.368 0.424 0.185 0.324Number of branddma 390 390 585 585 975 975Fixed effects are included at the brand-market and brand-year.Standard errors are clustered at the market level.** p<0.01, * p<0.05

performance gains in high Super Bowl viewership markets.

Table 4 begins by considering the same analysis in the pre-Super Bowl, w ≤ 0, time period.

Specification (1) illustrates that Budweiser earns more revenue per household in markets that will

eventually watch the Super Bowl. Some of these purchases may be in planning to watch the game,

while others are purchases to watch the same teams in a previous round of the playoffs. The effect is

slightly reduced by the inclusion of covariates in (2). (3) and (4) repeat the analysis for non-Budweiser

brands. These brands also exhibit increases in revenue per household in pre-Super Bowl weeks, though

insignificant. Budweiser’s greater revenue for consumption during the game may be the result of their

long-standing association with the Super Bowl.

Recalling the results from the previous subsection, it must be that the non-advertising brands’

effects of Super Bowl viewership terminate upon completion of the Super Bowl and the NFL season,

as opposed to the effects for the advertiser which persist. We test for this Pre vs. Post change in

specifications (5) and (6) without and with marketing variables respectively. The non-advertising

brands exhibit a significant loss (-0.05) of their revenue effect in the post-Super Bowl weeks, whereas

18

Budweiser does not exhibit any greater loss, despite having pre-Super Bowl performance that was more

than twice as strong as the non-advertisers. The -0.057 post-Super Bowl effect combined with the 0.157

pre-Super Bowl effect exactly gives the 0.100 effect observed in (5) in Table 2. In (6), the post-Super

Bowl drop in Budweiser’s performance relationship with ratings is insignificant, while the other brands

still realize a significant drop absorbing most of their 0.053 gain from (4).

Table 4: Effects of Super Bowl Viewership and Advertising in Beer

5 Pre and 8 Post Super Bowl Weeks Included(1) (2) (3) (4) (5) (6)

VARIABLES Bud Only Bud Only Non-Bud Non-Bud All Brands All BrandsPost-Super BowlRatings * Ad * w>0 -0.057* -0.043

(0.025) (0.023)Ratings * w>0 -0.046** -0.042*

(0.017) (0.016)Pre-Super BowlRatings * Ad 0.211** 0.183* 0.157** 0.149**

(0.074) (0.076) (0.048) (0.048)Ratings 0.054 0.053 0.054 0.049

(0.042) (0.041) (0.041) (0.042)MarketingGRPs 1.288** 0.035 0.529**

(0.092) (0.083) (0.046)Pre-SB NFL GRPs 0.003 -0.036 -0.006

(0.029) (0.025) (0.019)Price -8.940** -2.020** -2.352**

(1.471) (0.256) (0.299)Price * OtherBrands 4.480** 2.903**

(1.034) (0.907)Feature 0.248** 0.044** 0.053**

(0.061) (0.011) (0.013)Display -0.076 0.005 0.012

(0.075) (0.013) (0.019)Observations 4,694 4,694 18,775 18,775 58,984 58,984R-squared 0.187 0.305 0.207 0.241 0.219 0.252Number of branddma 173 173 692 692 1730+ 1730+Fixed effects are included at the brand-market and brand-year.Standard errors are clustered at the market level.** p<0.01, * p<0.05. + indicates brand-market-PrePost FEs.

We report the pre-Super Bowl patterns for the soda category in Table 5. Specification (1) documents

that the advertising major soda brand also earns more revenue when consumers buy beverages leading

up to the game. The Pre-Super Bowl coefficient on Ratings * Ad is 0.041. As opposed to the null effect

in which this gain should die out following the Super Bowl, (2) documents the advertiser actually earns

significantly more following the game with a 0.036 increase. Summing these up, we obtain the 0.077

Post-Super Bowl effect of Ratings * Ad that was reported in Table 3. The increase rises even further

with the inclusion of covariates in (3). Specifications (4) and (5) restrict the analysis to the smaller

19

brands and document that they incur a significant Post-Super Bowl decrease in revenue per household,

relative to pre-game. Specification (6) combines both large and small brands and documents that

the advertiser still earns more pre-game, but incurs a significant increase post-game. However, the

post-game increase is lost completely if both Coke and Pepsi advertised in the Super Bowl that year.

Table 5: Effects of Super Bowl Viewership and Advertising in the Soda Category

5 Pre and 8 Post Super Bowl Weeks Included(1) (2) (3) (4) (5) (6)

VARIABLES Coke-Pepsi Coke-Pepsi Coke-Pepsi Not Coke-Pepsi Not Coke-Pepsi All BrandsPost-Super BowlRatings * Ad * w>0 0.036* 0.041** 0.003 0.003 0.024**

(0.014) (0.013) (0.003) (0.003) (0.007)Ratings * Ad*Ad * w>0 -0.028** -0.035** -0.025**

(0.009) (0.008) (0.006)Ratings * w>0 -0.030 -0.023 -0.041* -0.034* -0.028

(0.016) (0.016) (0.017) (0.017) (0.017)Pre-Super BowlRatings * Ad 0.041** 0.041** 0.047** 0.010 0.003 0.020**

(0.012) (0.012) (0.011) (0.006) (0.006) (0.007)Ratings * Ad * Ad -0.007 -0.007 -0.016 0.002

(0.013) (0.013) (0.011) (0.011)Ratings -0.029 -0.029 -0.041 -0.010 -0.008 -0.017

(0.035) (0.035) (0.035) (0.044) (0.045) (0.040)MarketingGRPs -0.050 -0.297** -0.105

(0.056) (0.106) (0.057)Pre-SB NFL GRPs -0.281 0.036 -0.107

(0.208) (0.068) (0.113)Price -15.473** -4.740** -9.168**

(0.930) (0.366) (0.587)Price * OtherBrands -0.688 3.862*

(1.597) (1.588)Feature 0.132** 0.091** 0.163**

(0.030) (0.018) (0.026)Display -0.020 0.033* -0.011

(0.033) (0.015) (0.021)Observations 10,588 26,620 26,620 39,930 39,930 66,550R-squared 0.055 0.052 0.317 0.377 0.432 0.336Number of branddma 390 780+ 780+ 1170+ 1170+ 1950+Fixed effects are included at the brand-market and brand-year..Standard errors are clustered at the market level.** p<0.01, * p<0.05. + indicates brand-market-PrePost FEs.

One might be tempted to infer that the post-Super Bowl Ratings * Ad * w>0 coefficients should

be significantly greater than the pre-Super Bowl Ratings * Ad coefficients. It is however important

to recognize that a structural shift happens after the Super Bowl that should eliminate the pre-Super

Bowl patterns: the NFL season ends and the demand for consumption during games disappears. The

structural shift in consumption is clearly evident for the non-advertising beer brands in Table 4 and

the soda brands realize a drop in revenue per household in Super Bowl markets when they are not

advertising as evidenced by the negative Ratings * w>0 coefficients. Advertising either limits this

20

structural drop in performance or, in the case of soda, actually leads to even greater performance in

high Super Bowl viewership markets following the Super Bowl.

3.4 Return on Investment Calculations

It is tempting to extrapolate the marginal effect of ratings on revenue to a return on investment, but

such numbers are complicated by a variety of factors. On the return side, we must assume a margin for

the brand and scale our estimates to the national level.14 Scaling our estimates to the national level

is tricky. We observe a large, but non-random set of stores whose selection we are forced to ignore.15

We use national consumption numbers as a benchmark, but we must account for the fact that there

are many sales through bars, restaurants and other venues. A Super Bowl may shift some marginal

consumption in these locales, but in many cases, it may require convincing the management to switch

from Coke to Pepsi in the fountain. This suggests our grocery store substitution is likely not a good

proxy for out of store purchases of soda. Nevertheless, we assume that the effects reported above hold,

as a percentage of our observed industry revenue or profits, to the be same percent of actual total

industry profit as reported in the press.16

Computing the investment in Super Bowl advertising also requires assumptions. We do not know

the prices paid to air the ads, which could vary based on whether a competitor is advertising or not.

We assume the cost is $3m which is the typical price reported in the media during the time of our

analysis. We also assume a hypothetical one million dollar production cost of the ad. Together this

constitutes a $4m cost per ad, or a total $16m outlay for the 4 ads Pepsi typically ran and $36m for

the 9 ads we observed Budweiser airing in a single year. Noting these caveats, we calculate the return

on investment in the beer and soda categories.

The return on investment calculation for Budweiser is easiest as there are no concerns about com-

petitors advertising. Using specification (6) from Table 2, we multiply the Ratings * Ad coefficient of

0.105 by the average Super Bowl ratings of 0.45 to obtain an increase in observed revenue per household14A Harvard Business School case, “Cola Wars Continue: Coke and Pepsi in 2010”, suggests brands receive about 49

cents on each retail dollar.15For example, two of the largest retailers, Walmart and Costco, are not included in the sales data. To scale up, we

must assume customers at these stores respond in the same way to Super Bowl advertising as grocery store customers.16http://www.theatlantic.com/business/archive/2014/04/the-state-of-american-beer/360583/ for beer and

http://www.beverage-digest.com/pdf/top-10_2014.pdf for soda

21

of 0.047. This is 14.7 percent of Budweiser’s average weekly revenue as reported in Table 1. Applying

this percentage revenue increase to the $1.2b Budweiser earns in an eight week period, then multiplying

by a 50 percent gross margin, Bud would earn an extra $91m in profit at an investment cost of $36m

for the advertising, suggesting an ROI of 153 percent. This number is likely an overestimate because

our estimates are “local” to viewership variation in the 30 to 60 percent range. In other words, we

never observe the die-hard Super Bowl fans, who watch every year, not being exposed to a Budweiser

ad. Our analysis focuses on a more fickle Super Bowl viewer who may also be more fickle with regard

to other preferences and consequently more easily swayed by advertising.

In the soda category, we present the profit of each competitive scenario in a two-by-two in Table

6.17 If neither were to advertise, Coke would earn $1.647b of profit over 8 weeks, while Pepsi would

earn $1.182b. We construct this counterfactual assuming Super Bowl ratings equal zero. Advertising

by either increases profit by $27m, implying a return on the $16m investment of just over 165%.

However if both were to advertise, they would earn nearly $80m less than if neither had advertised.

Both advertising is also a pure strategy equilibrium in that both Coke and Pepsi would advertise if

their competitor was committed to advertising, rather than incur the losses of sitting out and losing

post-Super Bowl sales to their competitor. This is suggestive of a prisoner’s dilemma, but this finding is

driven by the negative ratings coefficient which is not statistically significant. If we set that coefficient

to zero and redid the analysis, both advertising would increase the profits relative to neither advertising.

22

Table 6: Two By Two Super Bowl Entry Comparison

Back of the Envelop 8 Week ProfitCoke Out Coke In

Pepsi Out 1.647 1.6741.182 1.060

Pepsi In 1.209 1.5681.525 1.103

Coke’s Profit listed above Pepsi’s in billions of $

Table 7: Testing Super Bowl Ad Viewership Interaction with NCAA Viewership

Pre and Post(20) Super Bowl Weeks Included(1) (2) (3) (4) (5) (6) (7) (8)

VARIABLES All Beer All Beer All Beer All Beer Coke Pepsi Coke Pepsi Coke Pepsi Coke PepsiPost-Super BowlRatings * Ad * w>0 0.096* 0.105* -0.062* -0.056 0.036** 0.052** -0.005 0.004

(0.044) (0.044) (0.030) (0.029) (0.009) (0.011) (0.013) (0.010)Ratings * Ad*Ad * w>0 0.010 -0.007 0.017 0.011

(0.014) (0.010) (0.009) (0.008)Ratings * w>0 0.005 0.003 -0.049** -0.043** -0.071 -0.076 -0.043 -0.033

(0.038) (0.038) (0.017) (0.016) (0.016) (0.042) (0.025) (0.030)Ratings * Ad * NCAA 0.544** 0.558** 0.544** 0.585** 0.729** 0.372** 0.729** 0.385**

(0.097) (0.097) (0.097) (0.098) (0.131) (0.105) (0.131) (0.105)Ratings * Ad*Ad * NCAA -0.657** -0.443** -0.657** -0.453**

(0.101) (0.085) (0.101) (0.085)Ratings * NCAA 0.144 0.086 0.144 0.092 -0.107 0.121 -0.107 0.119

(0.081) (0.080) (0.081) (0.079) (0.107) (0.089) (0.107) (0.089)NCAA * Bud -0.328** -0.347** -0.328** -0.371**

(0.041) (0.042) (0.041) (0.042)NCAA * Coke 0.214** 0.137**

(0.032) (0.027)NCAA * Pepsi -0.214** -0.124**

(0.032) (0.027)NCAA -0.214** -0.172** -0.214** -0.178** 0.039 0.011 0.253** 0.139**

(0.026) (0.025) (0.026) (0.024) (0.034) (0.032) (0.032) (0.029)Pre-Super BowlRatings * Ad 0.157** 0.159** 0.041** 0.047**

(0.048) (0.049) (0.012) (0.011)Ratings * Ad -0.007 -0.016

(0.013) (0.011)Ratings 0.054 0.047 -0.029 -0.043

(0.042) (0.042) (0.035) (0.037)MarketingGRPs -0.139 0.203* -0.175** -0.074

(0.149) (0.098) (0.057) (0.053)Pre-SB NFL GRPs 0.003 0.000 -0.320 -0.285

(0.018) (0.018) (0.209) (0.205)Price -2.818** -2.826** -16.607** -16.527**

(0.357) (0.343) (0.981) (0.973)Price * OtherBrands 4.030** 4.313**

(1.096) (1.047)Feature 0.063** 0.063** 0.139** 0.136**

(0.013) (0.013) (0.032) (0.031)Display 0.058* 0.046 0.011 0.012

(0.025) (0.024) (0.041) (0.039)Observations 88,795 88,795 112,264 112,264 40,080 40,080 50,668 50,668R-squared 0.148 0.178 0.156 0.187 0.039 0.307 0.041 0.312Number of branddma 865 865 1730+ 1730+ 390 390 780+ 780+Fixed effects are included at the brand-market-week and brand-year..Standard errors are clustered at the market level.** p<0.01, * p<0.05. + indicates brand-market-Pre-0-Post FEs.

23

3.5 Advertising and the Complementary Between Brands and Sports

To provide an explicit test for the relationship between Super Bowl ad effectiveness and subsequent

sports viewership, we collected data on viewership of the NCAA basketball tournament, which can

span across weeks 4 through 10, depending on the year. We chose to focus on the NCAA tournament

for two reasons. First, it occurs the soonest after the game, such that the effects have the potential

to be the greatest. Second, its viewership is spread more across network television than Major League

Baseball, whose teams primarily air their games on local cable networks. The AdViews data from

which we extracted sports viewership only reports ratings information for broadcasts on traditional

networks such as ABC, CBS, NBC and Fox.

To include the effects of NCAA viewership in our model, we use the entire 20 weeks of post-Super

Bowl data. We extend the post-Super Bowl group of coefficients to also include the NCAA viewership

variable by itself, then interacted with the major brands in the category, and finally interacted with

all of the Super Bowl ratings related coefficients from above.

Table 7 reports the interaction of Ratings * Ad with NCAA viewership in the fourth row. It is

significantly positive across all specifications which do or do not include covariates and which do or

do not include pre-Super Bowl weeks in the analysis. In the soda category, we observe that both

competitors advertising in the Super Bowl eliminates this complementarity between Super Bowl ad

exposure and the NCAA viewership just as it reduced the post-Super Bowl effect of the Ratings * Ad

coefficient. NCAA viewership itself tends to decrease revenue per household in beer, whereas Coke17The lower right box is the most commonly observed, so we set these profits to be equal to our estimate of 8 weeks

of each brand’s national profits. This is calculated by first taking 8/52’s of the reported $77.1b US soda industry andmultiplying it by an assumed 50% margin to obtain $11.862b of profit across 8 weeks. Next, we multiply this by eachbrand’s respective share of revenue as reported in Table 1 to obtain $1.568b and $1.103b in profits for Coke and Pepsirespectively. We assume the baseline is represented by a ratings of zero, then we need to subtract off the sum of theRatings, Ratings * Ad and Ratings *Ad*Ad coefficients from the numbers reported in the lower right. Using the resultsin (2) from Table 3, the sum of the three ratings coefficients in the post-Super Bowl scenario is -0.0340. Multiplying thisby the 50% margin and an average ratings of 45%, we get -0.0077. To scale this change up from our sample of stores,we assume the effect as a percent of the size of the market holds across our stores and the nation. The per-brand losswhen both brands advertise is thus 1.06% of the 8 week industry profit in our data. At the national level, this percentloss amounts to $63m per brand. Further removing the $16m spent on the advertising, the baseline profits for Coke andPepsi are $1.647b and $1.182b respectively. The loss when not advertising is based on the negative Ratings coefficientof -0.066 which is about 2 percent of the 8 week industry margin, or a loss of $122m. This reduces the non-advertisingprofits of Coke and Pepsi to $1.209b and $1.060b as represented in the bottom left and upper right boxes respectively. Ifthey advertise alone, they make $43m more because the Ratings * Ad coefficient multiplied by a 50% margin and 45%ratings is 2.78% of industry profits. Subtracting the $16m ad cost, the profits to Coke and Pepsi when advertising alonein the upper right and lower left boxes respectively are therefore $1.674b and $1.525b.

24

performs better if NCAA viewership is high and Pepsi performs worse.

Beverages such as soda and beer have long been consumed heavily during sporting events and

other occasions. These findings suggest that advertising plays a role in this complementarity and can

favor the advertiser’s brand over others. We find that eight weeks after the Super Bowl, the increased

advertising exposures in Super Bowl markets lead Super Bowl advertisers to earn more sales when

consumers purchase beverages to watch other sporting events. We are unaware of anyone else who has

documented this relationship, but it is in fact not surprising considering certain brands do garner a

much larger share of sales during sporting events.

3.6 Discussion and Caveats

There are some selection possibilities that deserve discussion. We conduct our analysis using variation

in ratings alone, conditional on a Super Bowl ad being aired. The two deviations in each soda brand’s

typical advertising decision are not likely candidates for a selection bias. Coca-Cola did not advertise

in the first year in our data (2006), but advertised in all subsequent years. As they had not advertised

since 1998, it does not appear they cherry-picked the year in our data in which they did not advertise.

Furthermore, the brand-year fixed effect would pick up any common demand shock to Coca-Cola in

2006 that might have led them to not advertise. The fixed effects therefore force selection concerns to

imply that a brand chose to advertise or not based on its expectations of the cross-market distribution of

Super Bowl viewership relative to that occurring in other years. Suppose the brand has little potential

to convert customers in politically left-leaning “blue” states, then it might withdraw from the Super

Bowl in a year when the competing teams will be from San Francisco and New York. This clearly

cannot describe Coca-Cola’s extended absence pre-2007 and persistence in the game ever since. Pepsi

on the other hand has only been absent in one year, 2010. There is however a lot of information about

this exit. Pepsi decided to shift both its Super Bowl budget, and a significant portion of the rest of

its marketing budget, to fund a social media campaign (the Pepsi Refresh campaign) that gave grants

to proposals to help local communities, the environment etc.. A Harvard Business School case and

the news articles it cites nicely describe this decision as about shifting emphasis to social causes and

not about a poor Super Bowl opportunity. In fact they announced the decision before the end of the

25

NFL’s regular season.18 It is therefore highly unlikely that this was driven by an accurate expectation

of which of the NFL teams would eventually make it to the Super Bowl. Pepsi did return the following

year when the Pepsi Refresh campaign failed to live up to Pepsi’s expectations.

It is possible that advertisers could alter other factors about their advertising in response to the

anticipated distribution of Super Bowl viewership. They could change the creative execution of the

ad, but the expense of developing creative for Super Bowls suggests this is unlikely in the weeks just

before the game. They could also alter the number of spots aired during the game. We can observe

this and have run specifications with this included, but prefer to focus our analysis around the ratings

data whose variation is exogenous. The brands could also alter the particular products they choose

to advertise in the game. Pepsi occasionally advertises Diet Pepsi or Pepsi Max, and Anheuser-Busch

has used some of its many spots in a year to include other brands such as Michelob or Stella. These

could have also been chosen strategically based on the anticipated distribution of viewership. While

we cannot rule out these selection decisions, we are skeptical they exist. They may bias upward the

estimates of the Super Bowl ad effect in beer, but its unlikely they account for a majority of it. In

soda, it is hard to imagine that such selection decisions would be driving the link between Super Bowl

viewership and sales in the particular weeks exhibiting spikes in Figure 4; and that these timed spikes

would only occur in years the soda brand is advertising.

It may also be argued that the post-Super Bowl performance of advertisers in high viewership

markets reflects a lasting effect from their greater consumption prior to the game. Such an effect could

be rationalized by habit persistence or switching costs, but i) such behavior is known to be difficult to

identify in aggregate data, ii) the above effects are much greater than others have found for packaged

goods (see Dube, Hitsch and Rossi, 2010), iii) this cannot explain why the revenue per household

increase in high viewership markets is actually greater after the Super Bowl in the soda category, and

iv) it is also inconsistent with some of the greatest effects occurring 5 or 8 weeks later.

Finally, there might be some concern about measurement error in the market-level ratings data.

This does not however seem to be a problem as we are finding large effects. We could have used the18A Wall Street Journal article titled “Pepsi Benches its Drinks” detailed the move on December 17, 2009, which is

three weeks before the end of the regular season.

26

Facebook likings of the competing teams as an instrument to remove such errors, but the first-stage

incremental fit is too small and greatly increases standard errors (see Rossi, 2014 for a discussion of

the problems with using weak instruments when there is little endogeneity concern). We have also

tested for measurement error by allowing coefficients to be interacted with number of households in

the market. Smaller markets in the data should exhibit more measurement error as there are fewer

local Nielsen panelists to form the share of household viewership numbers. We did not however find

this to effect the results.

4 Conclusion

Advertising is one of the most important instruments a brand can use to market its products, yet

advertising’s efficacy and the mechanisms behind it are highly disputed. The benefits for a new product

are clear as consumers might otherwise be unaware of its existence and features. For established brands,

the incentive to spend on uninformative advertising such as that observed in the Super Bowl has been

questioned. We document that there are large potential profits to such advertising.

The patterns of Super Bowl advertising profitability also uncovers the source of effectiveness and

competitive implications. Advertising can build or reinforce a complementarity between a brand and

potential consumption occasions. Consumption of beer, snacks and other items while watching sports

is easily observed in society, but this is, to our knowledge, the first evidence illustrating that advertising

plays a role in this relationship and can link the complementarity to a particular brand. Competition

for such a strong branding association is however shown to be dissipative and may result in a prisoner’s

dilemma.

27

References[1] Ackerberg, D. (2001), “Empirically Distinguishing Informative and Prestige Effects of Advertising,” RAND

Journal of Economics, 32(2), 316-333.

[2] Bagwell, K. (2007), “The Economic Analysis of Advertising,” Handbook of Industrial Organization, vol. 3,1701-1844.

[3] Becker, G. and K. Murphy (1993), “A Simple Theory of Advertising,” Quarterly Journal of Economics,108(4), 941-964.

[4] Bertrand, M., et.al. (2010), “What’s Advertising Content Worth? Evidence from a Consumer CreditMarketing Field Experiment,” The Quarterly Journal of Economics, 125(1), 263-306.

[5] Bronnenberg, B., M. Kruger and C. Mela (2008), “The IRI Marketing Data Set,” Marketing Science, 27(4),745-748.

[6] Bronnenberg, B., S. Dhar and J.P. Dubé, (2009), “Brand History, Geography, and the Persistence of BrandShares,” Journal of Political Economy, 117(1), 87-115.

[7] Crupi, A. (2012), “CBS Halfway to a Super Bowl Sellout,” AdWeek.

[8] Doganoglu, T. and D. Klapper (2006), “Goodwill and Dynamic Advertising Strategies,” Quantitative Mar-keting and Economics, 4(1), 5-29.

[9] Doraszelski, U. and S. Markovich (2007), “Advertising Dynamics and Competitive Advantage,” Rand Jour-nal of Economics, 38(3), 557-592.

[10] Dubé, J. P., G. Hitsch and P. Manchanda (2005), “An Empirical Model of Advertising Dynamics,” Quan-titative Marketing and Economics, 3(2), 107-144.

[11] Dubé, J. P., G. Hitsch and P. Rossi (2010), “State Dependence and Alternative Explanations for ConsumerInertia,” RAND Journal of Economics, 41(3), 417-445.

[12] Hendel, I. and A. Nevo (2003), “The Post-Promotion Dip Puzzle: What do the Data Have to Say?”Quantitative Marketing and Economics, 1(4), 409-424.

[13] Lewis, R. and J. Rao (2014), “The Unfavorable Economics of Measuring the Returns to Advertising,”working paper.

[14] Lewis, R. and D. Reiley (2013), “Down to the Minute Effect of Super Bowl Advertising on Online SearchBehavior,” 14th ACM Conference on Electronic Commerce 9(4) Article 61.

[15] Lodish, L., M. Abraham, S. Kalmenson, J. Livelsberger, B. Lubetkin, B. Richardson, and M. E. Stevens(1995), “How T.V. Advertising Works: A Meta-Analysis of 389 Real World Split Cable T.V. AdvertisingExperiments,” Journal of Marketing Research, 32(2), 125-139.

[16] Nerlove, M. and K.J. Arrow (1962), “Optimal Advertising Policy Under Dynamic Conditions,” Economica,29(114), 129-142.

[17] Rossi, P.E. (2014), “Even the Rich Can Make Themselves Poor: a critical examination of the use of IVmethods in marketing,” Marketing Science, 33(5), 655-672.

[18] Sahni, N. (2013), “Effect of Temporal Spacing Between Advertising Exposures: Evidence from an OnlineField Experiment,” working paper.

28