Super Bowl Ads∗
Wesley R. Hartmann
Graduate School of Business
Stanford University
Daniel Klapper
School of Business & Economics
Humboldt University Berlin
March, 2015
First 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. The
event has the potential to create significant increases in “brand capital” because ratings average
over 40 percent of households and ads are a focal point of the broadcast. Furthermore, variation
in exposures is exogenous because a brand cannot choose how many impressions it receives in each
market. 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 whether
advertisers’ sales are affected accordingly. We run our analysis using Nielsen ratings and store level
sales data in the beer and soda categories. We find that Super Bowl ads can generate significant
increases in revenue per household. However, when two major soda brands both advertise, much
of this gain is lost. Exploring the mechanism behind the ad effects, we find that Super Bowl ads
build 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 Lanier
Benkard, 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 Science
Conference, the 5th Workshop on the Economics of Advertising and Marketing in Beijing, the 2013 Summer Institute in
Competitive Strategy and 2014 NBER IO Summer Institute. Corey Anderson and Chris Lion provided valuable research
assistance 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 revenue
1
2
http://sports.espn.go.com/nfl/news/story?id=4751415
Competition 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 promotional 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 product 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 a
3
Extrapolations 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 section
of the paper.
4
Some of the Super Bowl ads in our analysis focus on smaller sub-brands, such as when Diet Pepsi Max had recently
been rebranded to Pepsi Max, but the majority emphasize just the brand, or well established product offerings such as
Bud 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 StephensDavidowitz 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.
5
Our abstract describing the value of exogenous variation arising from changes in Super Bowl competitors for identifying advertising effects was first published in an abstract in the 2012 Marketing Science Conference program. The
authors can provide a copy of this upon request.
6
Ackerberg (2001) and others document informative advertising effects arising for new products.
7
Nielsen actually releases data for the top 55 DMAs, but because of entry and exit from the top 55, we have 56.
8
A 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 region
and the ability to increase the number of DMAs in our analysis from roughly 33 to 56. The Nielsen data also allows us
to 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 crossmarket 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.
9
We’ve run our analysis also coding this up as an advertisement and while the coefficients change some, the substantive
conclusion remains the same.
6
100.0
15.0
10.0
90.0
5.0
70.0
-5.0
-10.0
60.0
DMA Average Ratings
Ratings Difference from DMA Average
80.0
0.0
-15.0
50.0
-20.0
40.0
-25.0
30.0
-30.0
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. Among
10
While 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
60
Percent of DMA that Watched the Super Bowl
55
50
45
40
35
30
0
5
10
15
20
25
30
35
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 prior
11
To 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 substantially
12
The 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
Beer
Variable
Budweiser
Miller
Coors
Corona
Revenue per HH
0.322
0.114
0.091
0.057
volume 6 pk
0.077
0.028
0.021
0.008
price per oz
0.059
0.058
0.061
0.098
feature
0.050
0.050
0.053
0.052
display
0.054
0.047
0.049
0.053
GRPs / 100
0.007
0.004
0.002
0.001
Pre-SB NFL GRPs /100
0.164
0.058
0.095
0.009
Observations across 173 DMAs and 20 weeks. Some DMA-years missing.
DMAs with alcohold restrictions or negligible sales omitted.
Soda
Variable
Coke
Pepsi
Dr Pepper
Mtn Dew
Revenue per HH
0.381
0.268
0.111
0.144
volume 6 pk
0.209
0.158
0.060
0.077
price per oz
0.026
0.025
0.027
0.028
feature
0.093
0.105
0.021
0.041
display
0.082
0.089
0.069
0.077
GRPs / 100
0.007
0.007
0.003
0.002
Pre-SB NFL GRPs / 100
0.006
0.020
0.012
0.001
Observations across 191 DMAs and 20 weeks. Some DMA-years missing.
3
Other
0.332
0.074
0.061
0.024
0.024
0.004
0.026
Other
0.537
0.288
0.027
0.036
0.047
0.004
0.001
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 = α1 Ajy Rmy + α2 Ajy Aky Rmy + δRmy + Xjmyw β + γF E + 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 viewership
13
We have also run the analyses below using volume per household and find the same substantive findings and significance 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. γF E 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
0.35
Revenue per Household Attributable to Super Bowl Ad Viewership
0.30
0.25
0.20
0.15
0.10
0.05
0.00
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
-0.05
-0.10
-0.15
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.
Revenue per Household Attributable to Super Bowl Ad Viewership
0.35
0.25
0.15
0.05
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
-0.05
-0.15
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 separately 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 indicates 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 marketing 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. Nonadvertising 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
VARIABLES
Post-Super Bowl
Ratings * Ad
8 Weeks Post-Super Bowl Included
(1)
(2)
(3)
(4)
Bud Only
Bud Only
Non-Bud
Non-Bud
0.109
(0.063)
0.128
(0.065)
Ratings
Marketing
GRPs
Pre-SB NFL GRPs
Price
0.008
(0.035)
0.096
(0.069)
0.013
(0.021)
-7.679**
(1.521)
Price * OtherBrands
Feature
0.202**
(0.050)
Display
-0.029
(0.068)
Observations
7,104
7,104
28,411
R-squared
0.243
0.32
0.243
Number of branddma
173
173
692
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
0.010
(0.035)
0.291**
(0.045)
-0.033
(0.021)
-1.342**
(0.185)
0.352
(0.996)
0.025**
(0.007)
0.026
(0.015)
28,411
0.267
692
(5)
All Brands
(6)
All Brands
0.100*
(0.042)
0.008
(0.035)
0.105*
(0.043)
0.008
(0.036)
35,515
0.243
865
0.180**
(0.043)
-0.004
(0.017)
-2.024**
(0.307)
1.055
(0.973)
0.043**
(0.011)
0.021
(0.021)
35,515
0.269
865
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 advertisers 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
VARIABLES
Post-Super Bowl
Ratings * Ad
Ratings * Ad * Ad
Ratings
Marketing
GRPs
Pre-SB NFL GRPs
Price
(1)
Coke-Pepsi
0.077**
(0.013)
-0.036*
(0.015)
-0.059
(0.037)
8 Weeks Post-Super Bowl Included
(2)
(3)
(4)
Coke-Pepsi
Not Coke-Pepsi
Not Coke-Pepsi
0.089**
(0.014)
-0.057**
(0.012)
-0.066
(0.038)
0.012*
(0.006)
0.007
(0.006)
-0.051
(0.038)
-0.042
(0.039)
-0.242**
(0.063)
-0.368
(0.214)
-15.101**
(0.907)
Price * OtherBrands
Feature
0.139**
(0.033)
Display
-0.041
(0.035)
Observations
16,032
16,032
24,048
R-squared
0.05
0.305
0.368
Number of branddma
390
390
585
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
-0.173
(0.109)
0.077
(0.068)
-4.797**
(0.373)
0.633
(1.538)
0.098**
(0.020)
0.022
(0.015)
24,048
0.424
585
(5)
All Brands
(6)
All Brands
0.047**
(0.008)
-0.020
(0.013)
-0.054
(0.036)
0.044**
(0.007)
-0.026*
(0.011)
-0.046
(0.036)
40,080
0.185
975
-0.294**
(0.062)
-0.123
(0.114)
-9.063**
(0.575)
5.027**
(1.535)
0.168**
(0.028)
-0.027
(0.022)
40,080
0.324
975
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
VARIABLES
Post-Super Bowl
Ratings * Ad * w>0
5 Pre and 8 Post Super Bowl Weeks Included
(1)
(2)
(3)
(4)
Bud Only
Bud Only
Non-Bud
Non-Bud
Ratings * w>0
Pre-Super Bowl
Ratings * Ad
0.211**
(0.074)
0.183*
(0.076)
Ratings
Marketing
GRPs
Pre-SB NFL GRPs
Price
0.054
(0.042)
1.288**
(0.092)
0.003
(0.029)
-8.940**
(1.471)
Price * OtherBrands
Feature
0.248**
(0.061)
Display
-0.076
(0.075)
Observations
4,694
4,694
18,775
R-squared
0.187
0.305
0.207
Number of branddma
173
173
692
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.
0.053
(0.041)
0.035
(0.083)
-0.036
(0.025)
-2.020**
(0.256)
4.480**
(1.034)
0.044**
(0.011)
0.005
(0.013)
18,775
0.241
692
(5)
All Brands
(6)
All Brands
-0.057*
(0.025)
-0.046**
(0.017)
-0.043
(0.023)
-0.042*
(0.016)
0.157**
(0.048)
0.054
(0.041)
0.149**
(0.048)
0.049
(0.042)
58,984
0.219
1730+
0.529**
(0.046)
-0.006
(0.019)
-2.352**
(0.299)
2.903**
(0.907)
0.053**
(0.013)
0.012
(0.019)
58,984
0.252
1730+
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
VARIABLES
Post-Super Bowl
Ratings * Ad * w>0
5 Pre and 8 Post Super Bowl Weeks Included
(1)
(2)
(3)
(4)
Coke-Pepsi
Coke-Pepsi
Coke-Pepsi
Not Coke-Pepsi
Ratings * Ad*Ad * w>0
Ratings * w>0
Pre-Super Bowl
Ratings * Ad
Ratings * Ad * Ad
Ratings
Marketing
GRPs
Pre-SB NFL GRPs
Price
0.041**
(0.012)
-0.007
(0.013)
-0.029
(0.035)
0.036*
(0.014)
-0.028**
(0.009)
-0.030
(0.016)
0.041**
(0.013)
-0.035**
(0.008)
-0.023
(0.016)
0.041**
(0.012)
-0.007
(0.013)
-0.029
(0.035)
0.047**
(0.011)
-0.016
(0.011)
-0.041
(0.035)
(5)
Not Coke-Pepsi
(6)
All Brands
0.003
(0.003)
0.003
(0.003)
-0.041*
(0.017)
-0.034*
(0.017)
0.024**
(0.007)
-0.025**
(0.006)
-0.028
(0.017)
0.010
(0.006)
0.003
(0.006)
-0.010
(0.044)
-0.008
(0.045)
0.020**
(0.007)
0.002
(0.011)
-0.017
(0.040)
39,930
0.377
1170+
-0.297**
(0.106)
0.036
(0.068)
-4.740**
(0.366)
-0.688
(1.597)
0.091**
(0.018)
0.033*
(0.015)
39,930
0.432
1170+
-0.105
(0.057)
-0.107
(0.113)
-9.168**
(0.587)
3.862*
(1.588)
0.163**
(0.026)
-0.011
(0.021)
66,550
0.336
1950+
-0.050
(0.056)
-0.281
(0.208)
-15.473**
(0.930)
Price * OtherBrands
Feature
Display
0.132**
(0.030)
-0.020
(0.033)
26,620
0.317
780+
brand-year..
Observations
10,588
26,620
R-squared
0.055
0.052
Number of branddma
390
780+
Fixed effects are included at the brand-market and
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 competitors 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 household
14
A Harvard Business School case, “Cola Wars Continue: Coke and Pepsi in 2010”, suggests brands receive about 49
cents on each retail dollar.
15
For 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.
16
http://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 Profit
Coke Out
Coke In
Pepsi Out
1.647
1.674
1.182
1.060
Pepsi In
1.209
1.525
1.568
1.103
Coke’s Profit listed above Pepsi’s in billions of $
Table 7: Testing Super Bowl Ad Viewership Interaction with NCAA Viewership
VARIABLES
Post-Super Bowl
Ratings * Ad * w>0
(1)
All Beer
Pre and Post(20) Super Bowl Weeks Included
(2)
(3)
(4)
(5)
All Beer
All Beer
All Beer
Coke Pepsi
0.096*
(0.044)
0.105*
(0.044)
-0.062*
(0.030)
-0.056
(0.029)
0.005
(0.038)
0.544**
(0.097)
0.003
(0.038)
0.558**
(0.097)
-0.049**
(0.017)
0.544**
(0.097)
-0.043**
(0.016)
0.585**
(0.098)
0.144
(0.081)
-0.328**
(0.041)
0.086
(0.080)
-0.347**
(0.042)
0.144
(0.081)
-0.328**
(0.041)
0.092
(0.079)
-0.371**
(0.042)
Ratings * Ad*Ad * w>0
Ratings * w>0
Ratings * Ad * NCAA
Ratings * Ad*Ad * NCAA
Ratings * NCAA
NCAA * Bud
NCAA * Coke
(6)
Coke Pepsi
(7)
Coke Pepsi
(8)
Coke Pepsi
0.036**
(0.009)
0.010
(0.014)
-0.071
(0.016)
0.729**
(0.131)
-0.657**
(0.101)
-0.107
(0.107)
0.052**
(0.011)
-0.007
(0.010)
-0.076
(0.042)
0.372**
(0.105)
-0.443**
(0.085)
0.121
(0.089)
-0.005
(0.013)
0.017
(0.009)
-0.043
(0.025)
0.729**
(0.131)
-0.657**
(0.101)
-0.107
(0.107)
0.004
(0.010)
0.011
(0.008)
-0.033
(0.030)
0.385**
(0.105)
-0.453**
(0.085)
0.119
(0.089)
0.214**
(0.032)
0.137**
(0.027)
-0.214**
(0.032)
0.253**
(0.032)
-0.124**
(0.027)
0.139**
(0.029)
0.041**
(0.012)
-0.007
(0.013)
-0.029
(0.035)
0.047**
(0.011)
-0.016
(0.011)
-0.043
(0.037)
NCAA * Pepsi
NCAA
Pre-Super Bowl
Ratings * Ad
-0.214**
(0.026)
-0.172**
(0.025)
-0.214**
(0.026)
-0.178**
(0.024)
0.157**
(0.048)
0.159**
(0.049)
0.054
(0.042)
0.047
(0.042)
0.039
(0.034)
0.011
(0.032)
Ratings * Ad
Ratings
Marketing
GRPs
-0.139
0.203*
(0.149)
(0.098)
Pre-SB NFL GRPs
0.003
0.000
(0.018)
(0.018)
Price
-2.818**
-2.826**
(0.357)
(0.343)
Price * OtherBrands
4.030**
4.313**
(1.096)
(1.047)
Feature
0.063**
0.063**
(0.013)
(0.013)
Display
0.058*
0.046
(0.025)
(0.024)
Observations
88,795
88,795
112,264
112,264
R-squared
0.148
0.178
0.156
0.187
Number of branddma
865
865
1730+
1730+
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
40,080
0.039
390
-0.175**
(0.057)
-0.320
(0.209)
-16.607**
(0.981)
-0.074
(0.053)
-0.285
(0.205)
-16.527**
(0.973)
0.139**
(0.032)
0.011
(0.041)
40,080
0.307
390
0.136**
(0.031)
0.012
(0.039)
50,668
0.312
780+
50,668
0.041
780+
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 Coke
17
The 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 and
multiplying it by an assumed 50% margin to obtain $11.862b of profit across 8 weeks. Next, we multiply this by each
brand’s respective share of revenue as reported in Table 1 to obtain $1.568b and $1.103b in profits for Coke and Pepsi
respectively. We assume the baseline is represented by a ratings of zero, then we need to subtract off the sum of the
Ratings, Ratings * Ad and Ratings *Ad*Ad coefficients from the numbers reported in the lower right. Using the results
in (2) from Table 3, the sum of the three ratings coefficients in the post-Super Bowl scenario is -0.0340. Multiplying this
by 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 loss
when both brands advertise is thus 1.06% of the 8 week industry profit in our data. At the national level, this percent
loss amounts to $63m per brand. Further removing the $16m spent on the advertising, the baseline profits for Coke and
Pepsi are $1.647b and $1.182b respectively. The loss when not advertising is based on the negative Ratings coefficient
of -0.066 which is about 2 percent of the 8 week industry margin, or a loss of $122m. This reduces the non-advertising
profits of Coke and Pepsi to $1.209b and $1.060b as represented in the bottom left and upper right boxes respectively. If
they 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 alone
in 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 the
18
A 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 Credit
Marketing 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 Brand
Shares,” 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 Marketing and Economics, 4(1), 5-29.
[9] Doraszelski, U. and S. Markovich (2007), “Advertising Dynamics and Competitive Advantage,” Rand Journal of Economics, 38(3), 557-592.
[10] Dubé, J. P., G. Hitsch and P. Manchanda (2005), “An Empirical Model of Advertising Dynamics,” Quantitative Marketing and Economics, 3(2), 107-144.
[11] Dubé, J. P., G. Hitsch and P. Rossi (2010), “State Dependence and Alternative Explanations for Consumer
Inertia,” 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 Search
Behavior,” 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. Advertising
Experiments,” 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 IV
methods in marketing,” Marketing Science, 33(5), 655-672.
[18] Sahni, N. (2013), “Effect of Temporal Spacing Between Advertising Exposures: Evidence from an Online
Field Experiment,” working paper.
28