Demand for “Healthy” Products: False claims in
advertising∗
Anita Rao
Emily Wang
Booth School of Business
Department of Resource Economics
University of Chicago
University of Massachusetts
April 2015
First draft: January 2015
Abstract
Firms often make selective or deceptive claims in their advertising. Such claims
can have negative consequences for consumers especially if consumers are not fully
informed and the claims are hard to verify. This paper aims to measure the impact
of such false claims on consumer demand and understand which type of consumer is
most impacted by these claims. Using a panel dataset of consumer purchases and firm
advertising at the ad-campaign-title level, we exploit the fact that four popular products settled charges raised by the Federal Trade Commission, leading to an exogenous
discontinuation of the false advertising campaigns, to measure this impact. We further
document firm responses in terms of price and advertisement changes around the date
of the warning letter.
∗
We thank the James M. Kilts Center for Marketing at the University of Chicago Booth School of Business
for their help with the Homescan, RMS and Media data. Thanks to seminar participants at Duke, NCSU; to
participants at the 2014 Marketing Science Conference and 2014 Marketing Dynamics Conference for their
helpful comments. Email addresses: [email protected], [email protected]
1
1
Introduction
Firms often have incentives to make deceptive or selective claims: especially when they
can lead to an increase in demand or can prevent substitution away from their products.
Examples abound in our day-to-day world: products selectively claiming to be made with
whole wheat even when whole wheat is not a main ingredient; products claiming to be “allnatural” even when they contain synthetic compounds. While regulatory bodies such as
the FTC exist to safeguard consumers from deception, the vast number of companies and
advertisements suggests that some claims are likely to go unnoticed for prolonged durations.
This paper asks the question, what (if any) impact do false claims have on consumers’
purchase decisions. While the role of information on consumer decisions has been studied1 ,
the consequences of false or misleading health claims on consumer purchases has received
very little empirical attention.2 If consumers are completely informed or these claims are
easily verifiable, any additional information without product composition change should have
no impact on demand. For example, if consumers know that a product does not have sugar
but contains other forms of sugar such as fruit juice concentrate, adding a “no added sugar”
label should have no impact on demand. However, if consumers do not know that fruit juice
concentrate is another form of sugar, this new claim can mislead consumers into thinking this
is a healthy product and can increase demand. Similarly, if these claims are easily verifiable
(e.g. through reading the ingredient list carefully) through search or experience (Stigler 1961,
Nelson 1970) then these claims should have no impact on demand. However, if these claims
are hard to verify then these can mislead consumers.
To measure this impact, we exploit the fact that a few companies were under investigation
by the FTC for making false health claims. These companies reached an agreement with
the FTC which required an immediate termination of the false claims. These agreements
are issued as a consent order with an accompanying press-release statement by the FTC. We
use the timings of these consent orders as exogenous shocks that a) reduced the levels of the
false advertising campaigns to 0 and b) led to widespread information diffusion about these
misleading claims.
As an example, Kellogg’s Frosted Mini-Wheats started making the claim in January 2008
1
For example, Bronnenberg, Dube, Gentzkow and Shapiro (2013) find that experts in a certain category
are more likely to buy the generic versions while novices are more likely to buy the branded versions of
an otherwise homogenous product; Ippolito and Mathios (1990,1995) find that after a regulatory ban on
advertising health benefits was lifted, information acquisition became easier and more people were able to
purchase healthier products; Jin and Leslie (2003) find that a policy change requiring restaurants to display
hygiene quality grade cards resulted in consumers becoming more sensitive to this information.
2
Stated purchase intentions and consumer beliefs about brands making deceptive claims has been studied
in lab settings, e.g. Darke and Ritchie (2007), Burke et al (1988), Dyer and Kuehl (1978) and Johar (1995).
2
that FMW was “clinically shown to improve kids’ attentiveness by nearly 20%” without any
change in the composition of their product. In April 2009, the FTC issued a consent decree
requiring Kellogg’s to stop making these claims. For Frosted Mini-Wheats this implied both
a discontinuation of television advertisements and change of their front-of-the-box labeling.
Our empirical strategy is based on comparing market shares before and after these FTC
issued public statements to give us a measure of the impact of these false claims.
Using individual-level purchase data from products across 4 categories which were impacted by these false claims, we find that the termination of these claims led to a significant
decline in demand. Our findings further indicate that it is the less-loyal group of consumers
(consumers who, prior to the FTC press-release, purchased fewer units of the focal brand
as a share of their total purchases in that category) who are most impacted by the misleading health claims. These consumers are likely to be the least informed about the brand
and hence the most impacted by the “new” information presented in the advertisements and
front-of-the-box packaging.
Perhaps closest to our paper is Peltzman (1981) who studies the effects of FTC regulation
on the capital market, advertising expenditure and market share of the impacted brands.
Peltzman postulates that false advertisements should impact first-time buyers rather than
loyal buyers. However, data limitations at the time prevented Peltzman from investigating
heterogenous effects and controlling for prices. We build on this paper by incorporating
individual-level data at the brand-week-store level and provide evidence corroborating Peltzman’s model.
Broadly, this paper contributes to the literature on firm deception (e.g. Jin and Kato
2006 and Zinman and Zitzewitz 2013) and the literature on the influence of front-of-package
information on consumers (e.g. Moorman 1996, Roe et al 1999) and firms (e.g. Moorman et
al 2012) by measuring the impact of misleading health claims.
In the next section we describe the data and provide reduced form evidence on the impact
of misleading claims on consumer demand. In section 3 we describe the demand estimation
that controls for prices and the competitive environment to quantify the impact of these
misleading claims. Section 4 describes the results, Section 5 documents firm responses in
terms of price and advertisement changes around the date of the warning letter and Section
6 concludes.
2
Data
We use the Nielsen homescan data which contains households’ purchases at the UPC-date
level, the RMS data which contains the weekly price at the UPC-store level for participating
3
retailers and the Nielsen media data which contains the ad-spend, air-time and frequency
of campaigns at the creative-title level for each brand. The Homescan data consists of a
panel of about 40,000-60,000 households. Both the Homescan and Media data span the
years 2004-2012 while the RMS data spans 2006-2012.
We combine this with the dates that the FTC issued consent orders pertaining to various
companies in the RTE cereal, yogurt, yogurt-drink and nutritional supplements categories.
The relevant population of firms impacted by the FTC warning letters can be found at the
FTC website3 and consists of cases and proceedings classified under the mission of Consumer
Protection and the topic of Health Claims. As of November 2014 there were a total of 113
such cases. We restrict our attention to four of these cases based on the following inclusion
criteria (the number in parentheses pertain to the number of cases that do not meet the
listed criteria): 1. The cases should not pertain to internet scams or products sold only
online (44) 2. The case filing date should be after 2003 and before 2012 to be within the
timeframe of the Homescan data (27) 3. The cases should involve consumer product goods
(e.g. insurance, tanning services etc. were excluded) and be present in the Nielsen products
file (23) 4. The total number of observations across households and months should be at least
1,000, to ensure statistically meaningful measures (19). This leaves us with four products
which are sold exclusively in retail stores. The table below lists these products, the claims
they were making and the date they were asked to terminate these claims.
Table 1: Products asked to terminate misleading claims by the FTC
Product
Category
Claim
Consent Order
Kel Mini-Wheats
RTE cereal improve kids’ attentiveness by 18%
April 2009
Dannon Activia
Yogurt
help with slow intestinal transit
Dec 2010
Dannon Dan Active Yogurt drink helps strengthen your body’s defenses
Dec 2010
Airborne
Supplement offers guaranteed cold-fighting protection Aug 2008
The FTC typically conducts a private investigation of firms’ claims prior to the release
of the formal and public complaint. While the firm is made aware of this investigation,
consumers and members of the press have no knowledge of this investigation. Thus the
informational impact of the termination notice occurs only after the date of the consent order.
However, firms can choose to make changes (for the better) to their marketing activities prior
to the announcement.
To provide some preliminary evidence that false claims may have an impact on consumer
demand Figures 1 - 3 plots the aggregate market shares over time of the impacted products. The farthest vertical lines in these graphs indicate the date of the warning letter, the
3
http://www.ftc.gov/enforcement/cases-proceedings/advanced-search
4
preceding vertical lines (if any) correspond to the start of the misleading claims.
RTE cereal
Figure 1 plots the market shares of Frosted Mini Wheats over time. The farthest vertical line
in the graph corresponds to April 2009 when the FTC issued a consent decree to Kellogg’s to
stop making claims that eating the cereal increased children’s attentiveness by nearly 20%.
The plot indicates a fairly sharp decrease in market share after this event and a symmetric
increase in market share before this event. Figure 7 in Section 5.2 provides evidence that
the relevant advertisements were discontinued after the warning letter was sent.
Jul 2012
Jan 2013
Jul 2011
Jan 2012
Jul 2010
Jan 2011
Jul 2009
Jan 2010
Jan 2009
Jul 2008
Jan 2008
Jul 2007
Jan 2007
Jul 2006
Jan 2006
Jul 2005
Jan 2005
Jul 2004
Jan 2004
Jul 2003
.02
.03
Market share
.04
.05
.06
.07
Kel Frosted Mini-Wheats
Figure 1: Frosted Mini-Wheats market shares over time
Yogurt and Yogurt-drinks
In the refrigerated yogurts category, Dannon was issued a consent decree in December 2010
to stop making the claim that Activia “relieves irregularity” and DanActive “helps people
avoid catching colds or the flu”. These claims were present in the brands since introduction:
for Activia in February 2006 and DanActive in January 2007.
5
Dannon Dan Active
Figure 2: Dannon Activia (yogurt) and DanActive (yogurt-drink) market shares over time
Nutritional Supplements
Airborne is a dietary supplement that since its introduction claimed that it had “guaranteed
cold-fighting properties”, that “if taken at the first sign of a cold symptom, its herbal formulation is clinically proven to nip most colds in the bud”. The FTC in August 2008 issued
a consent order to the company to stop making these claims which were unsubstantiated.
Figure 3 shows a marked decrease in market shares following the FTC request of terminating
these claims.
Figure 3: Airborne market shares over time
6
Jan 2013
Jul 2012
Jan 2012
Jul 2011
Jan 2011
Jul 2010
Jan 2010
Jul 2009
Jan 2009
Jul 2008
Jan 2008
Jul 2007
Jan 2007
Jul 2006
Jan 2006
Jul 2005
Jan 2005
Jul 2004
Jan 2004
Jul 2003
0
.02
Market share
.04
.06
.08
Airborne
Jul 2012
Jan 2013
Jul 2011
Jan 2012
Jul 2010
Jan 2011
Jul 2009
Jan 2010
Jul 2008
Jan 2009
Jul 2007
Jan 2008
Jul 2006
Jan 2007
Jul 2005
Jan 2006
Jul 2004
Jan 2005
Jan 2004
Jul 2003
0
Jul 2013
Jul 2012
Jan 2013
Jan 2012
Jul 2011
Jul 2010
Jan 2011
Jul 2009
Jan 2010
Jan 2009
Jul 2008
Jan 2008
Jul 2007
Jul 2006
Jan 2007
Jan 2006
Jul 2005
0
.1
Market share
.2
Market share
.02
.04
.3
.4
.06
Dannon Activia
These figures indicate that there is a perceptible impact of the removal of false claims
and provide preliminary evidence to the extent that consumers might be impacted by these
misleading claims. However, it is possible that some unobservable trend which coincided
with the timing of these warning letters impacted the entire category. To rule out this
explanation, we perform a reduced-form regression of market shares on the time since the
warning letter. We compare these results to the same regression run on similar products but
which are unlikely to be substitutes for the impacted product.
2.1
Reduced-form regression
We estimate the following reduced-form regression for the brands impacted by the warning
letter and for control brands
sjt = αj +
αj1 .
t
X
I (False Claims f,τ ) +
τ =1
αj2
t
X
I (FTC f,τ )
τ =1
Here, sjt is the market share of brand j in month t,
t
P
I (False Claims f,τ ) is the cumulative
τ =1
number of months till t that focal brand f has been making the false claims and
t
P
I (FTC f,τ )
τ =1
is the cumulative number of months since the warning letter has been issued to focal brand
f.
In this specification, αj2 measures the change in market share relative to the market share
just before the warning. A negative αj2 indicates a drop in market share after the issuance
of the warning letter. If there is indeed an effect of the termination of the false claims
on demand for the impacted brand, we should see a significantly more negative coefficient
2
associated with αj=f
than with αj62=f .
Control brands
Comparing the impacted brands to other similar products can be problematic especially
if consumers substitute to/away from these similar products as a response to the warning
letters. To get around this problem, we use products which are similar to but are unlikely
substitutes for the impacted product.
In the RTE cereal category, we use Quaker Oats which is a hot cereal as a control.
Since this is not a RTE cereal but is still likely to be impacted by the same unobservable
trends as all cereals this serves as a good control. Similarly, for the yogurt and yogurt-drink
categories, we expand the market to include refrigerated puddings as well. We then use Swiss
7
Miss Puddings as the relevant control brand. In the nutritional supplements category, we
use Bausch & Lomb products which are vitamins meant only for eye health.
Table 2: Reduced-form evidence: Market shares of impacted brands drop after warning
letters
Impacted brand Control Brand
co-eff
t-stat
co-eff
t-stat
FMW
Quaker Oats
Cumulative False Claims
0.0007
7.34 -0.0005
-1.52
Time since FTC warning letter -0.0005
-6.03 -0.0001
-0.31
0.0294
42.44 0.0681
25.79
Airborne
Bausch and Lomb
Cumulative False Claims
0.0002
1.98 -0.0001
-4.21
Time since FTC warning letter -0.0006
-3.06 0.0000
0.20
0.0163
2.79 0.0213
28.8
Activia
Swiss Miss Pudding
Cumulative False Claims
0.0008
19.22 -0.00006
-11.48
Time since FTC warning letter -0.0022
-7.64 0.00003
0.54
0.0288
19.06 0.00662
41.53
Dan Active
Swiss Miss Pudding
Cumulative False Claims
0.0002
7.16 -0.0001
-10.38
Time since FTC warning letter -0.0011
-5.84 0.0000
0.75
0.0055
8.46 0.0063
41.2
Table 2 reports the results of this regression. The estimates indicate that all the impacted
brands face a decline in market share after the date of the issuance of the warning letter.
This is largely consistent with Figures 1-3 as well. Moreover, the estimates for the control
brands do not show any significant drop in shares after the respective dates for each of the
impacted brands.
These regressions provide some evidence that it is unlikely that the drop in market share
for the impacted brands is due to some other unobservable event which coincided with the
issuance of the warning letter. We now turn to a consumer choice model that controls for
prices and the competitive environment.
3
Model
Within a product category, we assume that a household makes a choice from C options
every time a transaction is made. The price of the chosen option is directly observed from
the purchase panel data. We construct the prices of the other options in the consumer’s
consideration set using the RMS data. We construct average within-DMA weekly prices
8
by averaging prices (per lb) of each brand across all stores and UPCs each week. We then
match a household to these DMA-level prices. For those DMA-weeks where this observation
is missing, we use the weekly national average (average across DMAs).
We specify individual i’s utility from purchasing brand j at time t as
Pre False
uijt (θ) =
During False Claims
After FTC
}|
{ z
}|
{
t
t
X
X
false
false
ftc
ftc
αij
I (Falseτ ) + βij
I (Falseτ ) +αij
I (FTCτ ) + βij
I (FTCτ ) + γi pjt + εijt
z
z}|{
pre
αij
+
τ =1
τ =1
pre
where αij
is consumer i’s preference for brand j before the focal brand began making
false
the false claim. αij
, βijfalse measure the intercept and slope during the false claims period.
ftc
, βijftc measure the intercept and slope in the period after the FTC press-release, i.e. once
αij
ftc
the consumer knows that the claim is false. βif
< 0 indicates declining demand for the
focal brand f after the withdrawal of the claim; |βfftc | > |βjftc | indicates that this decline
is the steepest for the focal brand f . γi is her price-sensitivity coefficient. θ is the set of
pre false false ftc ftc parameters αij
, αij , βij , αij , βij , γ governing a consumer’s decision. We estimate a
different specification for each category to capture the specifics of that category.
In the yogurt and yogurt-drinks category, there is no “before” period as the claims were
present since product introduction. Hence we use the following specification
uijt (θ) =
false
αij
I (Falseτ )+βijfalse
t
X
I (False
ftc
Claimsτ )+αij
I (FTCτ )+βijftc
τ =1
t
X
I (FTCτ )+γi pjt +εijt
τ =1
Lastly, in the nutritional supplements category, seasonality -whether it is the flu season
or not - is captured using the following specification
false,notFlu
ftc,notFlu
false
ftc
uijt (θ) = αij
I (Falseτ )+αij
I (notFlu*Falseτ )+αij
I (FTCτ )+αij
I (notFlu*FTCτ )+βi pjt +εijt
The probability that individual i chooses brand j at time t is then given by
P
pi,t (θ) =
j
euijt (θ) Ii,t (j)
P u
ijt
je
where Ii,t (j) are indicator functions reflecting individual i0 s choice in purchase occasion
t.
Aggregating the probabilities over all purchase occasions that i makes, the individual-level
9
probability is
pi (θ) =
T
Y
pi,t (θ)
t=1
The overall log-likelihood across all individuals can then be written as
LL (θ) =
N
X
log pi (θ)
i=1
We assume an individual’s choice set to be fixed. For cereal, we use the top 8 brands
by total sales and Store Brand Mini-Wheats. For yogurt drinks, we use the top 4 brands
excluding those that were discontinued. We aggregate the rest of the products into Other
brands and normalize the utility from Other brands to 0.
4
Results
The results of the demand estimation for all categories are presented in Tables 14 and 17.
Specification S2 controls for price of the purchased brand as well as prices of other brands
in the consumer’s consideration set. Both specifications S1 and S2 cluster at the household
level. Table 3 reports the coefficient βjftc for RTE cereal, yogurt and yogurt drinks. For
the RTE cereal and yogurt drink categories, βfftc < 0 indicating declining demand for the
focal brand f after the withdrawal of the claim. Moreover, |βfftc | > |βjftc | indicating that this
decline is the steepest for the focal brand f .
Table 3 also reports the difference in slopes, βjftc − βjfalse ; and in the case of Nutritional
supplements the difference in absolute levels during the flu season, αjftc − αjfalse . This provides
further evidence of the decline in demand after the FTC consent order. Except for the yogurt
category (which we investigate further in Section 4.1 ), all impacted brands face a significant
decline in demand even after controlling for price and the competitive environment.
Figure 4 plots the decline in purchase units 4 months after FMW was required to terminate the false claims relative to the market share just before. There are 128,359 households
in the purchase panel and 115 million households in the US. Scaling the decline in purchase
units to the population level, Kellogg’s is likely to have experienced a decline of about 14m
units 4 months after the FTC letter which at $3.00 per unit translates to a $43m decline
in revenues that month.Similarly Figure 5 plots the decline in market share for Dan Active.
Projecting this to the population, Dan Active is likely to have faced a decline of 143,000
units or an equivalent of $505,000 in revenues.
From consumer’s and a regulatory perspective, Kellogg’s agreed to a $4 million settlement
10
in May 2013. To evaluate the magnitude of this settlement, we compute the revenue gain
for Frosted Mini-Wheats due to the presence of the claims. For Frosted Mini-Wheats, if we
assume that the entire gain in market share from January 2008 (start of the false claims)
to September 2010 (when market shares seem to stabilize to pre-2008 levels) was due to the
presence of the false claims, the annual revenue gain from these claims is $462m. Even after
controlling for advertisement expenses (which were not that different compared to 2008), the
revenue gain is much larger than the $4m settlement reached.
11
Table 3: Demand Estimates: Decline after the FTC letter: βjftc ; Decline after FTC letter
relative to period before: βjftc − βjfalse
RTE Cereal
Brand
βjftc
t-stat βjftc − βjfalse
Kel Frosted Mini-Wheats -0.012 -31.75
G M Cheerios
0.026 56.55
G M Honey Nut Cheerios
0.061 106.26
Post Honey Bunches of Oats -0.005 -10.95
Kel Frosted Flakes
0.092 145.73
G M Cinnamon Toast Crunch 0.046 62.63
Kel Rice Krispies
0.016 21.93
Kel Raisin Bran
0.014 25.24
CTL Frosted Mini-Wheats
0.021 33.82
t-stat
-0.075
0.116
0.024
-0.068
0.015
0.030
0.037
-0.035
0.002
-55.33
48.20
11.01
-40.48
5.47
10.94
12.01
-13.34
0.86
t-stat βjftc − βjfalse
t-stat
Yogurt
Brand
Dannon Activia
Yoplait
Stonyfield
Dannon
Yoplait Whips!
Chobani
βjftc
0.038 107.90
0.015 69.97
0.051 76.15
0.030 99.86
0.000
0.84
0.031 139.81
0.034
88.29
0.011
42.84
0.047
69.87
0.005
14.51
-0.002
-4.86
-0.048 -113.48
Yogurt Drink
Brand
Dannon Dan Active
Dannon Danimals
Lifeway
Stonyfield
βjftc
t-stat βjftc − βjfalse
-0.026 -46.21
-0.016 -13.63
0.017 13.57
-0.002 -0.71
Nutr Supplements
Brand
Airborne
Nature Made
Nature’s Bounty
Emergen-C
-0.061 -106.88
-0.055 -46.55
-0.037 -27.97
0.012
3.58
αjftc − αjfalse
-0.714
-0.013
0.314
0.726
12
t-stat
t-stat
-91.43
-0.95
19.32
30.42
Figure 4: Decline in market share 4 months after termination of false claims is steepest for
Frosted Mini-Wheats
Figure 5: Decline in market share 4 months after termination of false claims is steepest for
focal brands: Dan Active and Airborne
4.1
Heterogeneity in consumer responses
We next explore heterogeneity in consumer responses to see who is most impacted by these
claims. Consumers who differed on observable demographics were found to exhibit little difference in purchase behavior. This finding is consistent with Rossi, McCulloch and Allenby
(1996) who found that purchase histories were more informative than observable demographics. We develop a measure of consumer loyalty as a function of purchase behavior prior to the
issuance of the FTC press-release statement. We capture loyalty as the ratio of the Number
of Focal brand purchases to the Number of All Other brand purchases in that category. We
then classify a household as loyal if she is in the 90th percentile of this measure across all
households.
Across all brands, where we measure the slope decline βjftc , the most loyal consumers
are the least impacted by the termination of the false claims. The drop in market share for
the infrequent purchasers is substantial indicating that the informational impact of the FTC
warning letter was highest on this group of consumers. For the case of Airborne, where we
13
directly measure the level change before and after the termination notice, we find that the
most loyal consumers’ market share drops more than the infrequent consumers’.
Table 4: Heterogeneity in βjftc : Loyal consumers less impacted
ftc
ftc
βNotLoyalj
− βLoyalj
Kel Frosted Mini-Wheats
G M Cheerios
G M Honey Nut Cheerios
Post Honey Bunches of Oats
Kel Frosted Flakes
G M Cinnamon Toast Crunch
Kel Rice Krispies
Kel Raisin Bran
CTL BR Frosted Mini-Wheats
Dannon Activia
Yoplait
Stonyfield
Dannon
Yoplait Whips!
Chobani
t-stat
-0.0714 -151.06
0.0178 22.34
0.0504 52.94
-0.0235 -45.75
0.0817 71.13
0.0391 31.63
0.0105
7.51
-0.0117 -15.73
0.0068
8.68
-0.13 -321.92
0.02 80.06
0.01
7.69
0.01 25.46
0.03 43.67
-0.005 -14.70
Dannon Dan Active
Dannon Danimals
Lifeway
Stonyfield
-0.169 -141.49
0.008
2.81
-0.034 -13.97
0.029
2.59
Table 5: Heterogeneity in αjftc : Loyal consumers less impacted
αftc − αfalse NotLoyalj − αftc − αfalse Loyalj t-stat
Airborne
1.404 84.28
Nature Made
-0.145 -3.24
Nature’s Bounty
-0.006 -0.11
Emergen-C
0.417 6.70
Our measure of the change in demand should be thought of as demand under the new
equilibrium regime where the impacted brand can no longer make these claims. In other
words, after the exogenous event firms (both focal and comeptitors) are likely to respond
via price changes or ad-spend changes and our measure is a cumulative account of consumer
responses to the termination of the claims as well as firms’ strategic responses to the event.
14
Our demand estimation accounts for strategic changes in competitor’s prices. We explore
firm-side changes further in the following section.
5
Firm Response
In this section, we explore possible firm-side responses in terms of price and advertisement
responses.
5.1
Prices
Figure 6 plots the price per lb of the impacted brands averaged across DMAs. With the
exception of Airborne, there appears to be little perceptible change in prices around the
time of the warning letter. Airborne faces an increase in prices starting about a year before
the letter followed by a decrease in prices. Further, confidence intervals are not tight (not
shown here) indicating substantial variation in dma-specific prices. We therefore turn to
exploring variation in DMA-level prices (averaged across all stores within DMA).
We run a regression with dma-level weekly prices for the focal and competitor brands
1
2
pjdt = βjd
+ βjd
.I(F T C) + g (t) + εjdt
where j is the brand, d is the DMA and t is the week. I(FTC) is 1 if the week is within
1
the 4-month period after the FTC letter and 0 otherwise. βjd
is the estimate of the average
2
price in a dma and βjd
measures the change in prices for the 4-month period after the FTC
warning letter. g (t) is a flexible-function of time that serves as a control for time-trends.
2
Tables 6-9 report, for each brand in each category, the percentage of DMAs where βjdt
is significantly positive or significantly negative and the minimum, maximum and median in
both cases. A few patterns are worth noting: 1. Other than in the cereal category, the focal
brand faces a significant drop in price in either the directly impacted brand (Dannon Activia
in Yogurt, Airborne in Supplements) or the sister brand (Dannon in Yogurt, Danimals in
Yogurt drinks) in greater than 30% of the DMAs. 2. Competitive response is strongest in
the Cereal category from the Store brand Mini-Wheats where the price drop is significantly
negative in over 50% of the DMAs following the warning letter.
15
Figure 6: Average Prices (per lb) of the impacted brands
Table 6: DMA-level price changes 4 months after FTC warning: RTE Cereal
Significantly % DMAs Min Max Median
Kel Frosted Mini-Wheats
G M Cheerios
G M Honey Nut Cheerios
Post Honey Bunches of Oats
Kel Frosted Flakes
G M Cinnamon Toast Crunch
Kel Rice Krispies
Kel Raisin Bran
CTL BR Frosted Mini-Wheats
+
+
+
+
+
+
+
+
+
16
35%
4%
14%
17%
16%
6%
34%
0%
11%
2%
5%
2%
16%
5%
0.5%
11%
5%
52%
$0.14
-$0.30
$0.42
-$1.05
$0.20
-$0.62
$0.20
–
$0.19
-$0.33
$0.17
-$0.25
$0.23
-$0.50
$0.13
-$0.22
$0.00
-$0.42
$0.30
-$0.15
$1.38
-$0.41
$0.36
-$0.22
$0.46
–
$0.32
-$0.20
$0.31
-$0.17
$2.22
-$0.23
$0.13
-$0.12
$1.75
$0.00
$0.20
-$0.19
$0.66
-$0.54
$0.25
-$0.27
$0.24
–
$0.25
-$0.24
$0.21
-$0.20
$0.38
-$0.30
$0.13
-$0.14
$0.18
-$0.15
Table 7: DMA-level price changes 4 months after FTC warning: Yogurt
Significantly % DMAs
Min Max Median
Dannon Activia
Yoplait
Stonyfiled
Dannon
Yoplait Whips!
+
+
+
+
+
-
7%
44%
17%
9%
15%
15%
7%
32%
4%
3%
$0.43
-$1.75
$0.17
-$0.71
$0.00
-$68.98
$0.52
-$1.88
$0.31
-$1.14
$1.41
-$0.42
$0.74
-$0.18
$2.77
-$0.03
$7.57
-$0.51
$0.70
-$0.33
$0.79
-$0.62
$0.28
-$0.21
$1.05
-$0.77
$0.84
-$0.79
$0.41
-$0.48
Table 8: DMA-level price changes 4 months after FTC warning: Yogurt Drinks
Significantly % DMAs Min Max Median
Dannon Dan Active
Dannon Danimals
Lifeway
Stonyfield
+
+
+
+
-
20%
5%
9%
32%
34%
14%
19%
11%
$1.11
-$2.68
$0.82
-$4.92
$0.01
-$0.76
$1.11
-$2.82
$7.52
-$0.88
$1.82
-$0.83
$28.20
$0.00
$7.52
-$0.88
$2.83
-$1.20
$1.22
-$1.25
$1.12
-$0.37
$2.68
-$2.19
Table 9: DMA-level price changes 4 months after FTC warning: Nutritional Supplements
Significantly % DMAs Min Max Median
Airborne
Nature Made
Nature’s Bounty
Emergen-C
5.2
+
+
+
+
-
16%
34%
23%
1%
13%
1%
16%
11%
$1.63
-$5.46
$0.17
-$0.25
$0.16
-$0.32
$0.82
-$2.52
$10.48 $2.42
-$1.67 -$2.44
$0.50 $0.23
-$0.21 -$0.23
$0.33 $0.18
-$0.25 -$0.29
$3.40 $1.00
-$0.85 -$1.16
Advertisements
Figure 7 plots the ad-spend by creative title for Kellogg’s Frosted Mini-Wheats from Jan.
2008 - Dec. 2009. The dotted lines indicate the advertisements relevant to the misleading
17
claims (all featuring children going to or in school) and the solid lines indicate other advertisements. The solid black line is the envelope of total ad-spend on Frosted Mini-Wheats.
While the two creatives which started in 2008 had already stopped, the ones that started
in early 2009 were stopped following the FTC warnings4 . The envelope of total ad spend
indicates that although the misleading ads were removed, the overall ad spend still remained
high. Figure 8 plots the total ad-spend across all creatives for the competitor brands: G M
Cheerios, G M Honey Cheerios and Post Honey Bunches of Oats. These figures indicate that
there was little change in total ad-spend by the focal firm or the competitor firms following
the FTC warning letter.
Figure 7: Frosted Mini-Wheats Ad Spend by creative-title
Figure 8: Competitor’s Ad Spend
To quantify possible changes in advertisemtns, we estimate the following regression equa4
The ad that restarted late 2009 had the same creative but with the voice over making much weaker
claims.
18
tions for ad-spend, duration and frequency of advertisements at the brand-week level.
2
Adjt = γjt + γjt
.I (FTC) + εjt
2
where Ad is the vector of ad-related variables {AdSpend, Duration, Frequency}and γjt
is the vector of coeffcients measuring any changes corresponding to the 4 month-window
after the FTC warning letter. The focal brand did not have a significant change in advertisements as measured by these three variables. In the cereal category, only General Mills
Cheerios, General Mills Cinnamon Toast and Post Honey Bunches of Oats exhibited a significant increase in advertisements in this period. At this aggregate-level, competitors in other
categories had no significant change in advertisements. One caveat in the current analysis is
that it aggregates any variation across markets and creative-titles.
19
Table 10: Advertising regressions: RTE Cereal
Ad Spend ($) Duration (sec) Frequency
Kel Frosted Mini-Wheats
FTC
N obs
G M Cheerios
FTC
N obs
Post Honey Bunches of Oats
FTC
N obs
Kel Frosted Flakes
FTC
N obs
G M Cinnamon Toast Crunch
FTC
N obs
Kel Rice Krispies
FTC
N obs
Kel Raisin Bran
FTC
N obs
20
29785
(1.03)
1272
4473
(1.20)
149
(1.20)
102196
(2.25)
1272
14216
(3.17)
391
(2.38)
55777
(1.03)
477
8855
(1.48)
851
(2.90)
19279
(0.76)
636
327
(0.15)
-22
(-0.28)
194456
(4.32)
159
12808
(2.53)
853
(2.63)
5759
(0.09)
318
1976
(0.19)
66
(0.19)
-24134
(-0.43)
477
3715
(0.68)
123
(0.68)
Table 11: Advertising regressions: Yogurt
Ad Spend ($) Duration (sec) Frequency
Dannon Activia
FTC
N obs
Yoplait
FTC
N obs
Stonyfield
FTC
N obs
Other Dannon
FTC
N obs
26877
(0.80)
1470
3369
(0.50)
324
(1.25)
15363
(1.20)
5439
2358
(1.48)
91
(1.31)
110
(0.06)
147
125
(0.61)
4
(0.61)
4680
(0.45)
1617
-1977
(-0.85)
-41
(-0.33)
Table 12: Advertising regressions: Yogurt drinks
Ad Spend ($) Duration (sec) Frequency
Dannon DanActive
FTC
N obs
Dannon Danimals
FTC
N obs
-8342
(-1.01)
1470
-2001
(-1.42)
-83
(-1.39)
-9579
(-1.61)
441
-1399
(-1.31)
-47
(-1.31)
Table 13: Advertising regressions: Nutritional Supplements
Ad Spend ($) Duration (sec) Frequency
Airborne
FTC
N obs
-770.338
-0.03
318
21
1208.833
0.38
113.1974
0.6
5.3
Availability
Following the FTC consent order, the impacted products might be out-of-stock in stores
(for e.g. manufacturers may need a few weeks to replace the packaging of their existing
products). Lack of availability of the impacted products can explain the patterns observed
in the data. To ensure that demand side factors, and not product stock-outs, drive the decline
in market shares following the FTC consent order, we check for discontinuity patterns in store
availability.
An observation in the RMS data can be missing if 1) the store did not report sales of the
UPC that week: This, if it occurs, should occur randomly and not systematically after the
FTC consent order, 2) the UPC had no sales in that store-week and 3) the product was not
available. While separating out (2) and (3) is hard given the data, we take advantage of the
fact that stock-outs, if they occur, should effect the entire brand and not just a single UPC.
A brand typically has 20-70 UPCs associated with it. Aggregating UPCs to the brand level,
Figure 9 plots the number of stores in which at least 1 unit of the brand was sold.
Across all brands, only Dannon’s DanActive exhibits a sharp drop in the count of stores
where at least 1 unit of the product was sold. We next verify if the demand patterns
documented thus far hold for DanActive, controlling for availability. To do this, we exclude
those stores where a product was available in early 2010, but unavailable late 2010 and early
2011. We restrict attention to only those stores present in the RMS data. The market shares
are remarkably close to the ones plotted in Figure 2 indicating that the decline in market
share is likely to be due to demand-side factors.
22
Figure 9: Number of Stores in Which At Least 1 unit of the Brand was Sold
Jul 2012
Jan 2013
Jul 2011
Jan 2012
Jul 2010
Jan 2011
Jul 2009
Jan 2010
Jul 2008
Jan 2009
Jul 2007
Jan 2008
Jul 2006
Jan 2007
Jul 2005
Jan 2006
Jul 2004
Jan 2005
Jan 2004
Jul 2003
0
.1
Market share
.2
.3
.4
Dannon Dan Active
Figure 10: Dannon DanActive Market Shares Excluding Stores where it was Likely Not
Available
23
6
Conclusion
This paper finds that false claims can have a significant impact on consumer demand from
the evidence that termination of these claims led to a decline in consumer demand. While
introduction of these claims in firms’ packaging and advertising messages is likely to be
endogenous, the termination of these claims was an exogenous event driven by the efforts of
the FTC. We exploit this exogenous event and measure market shares and individual-level
purchase behaviors before and after this event controlling for time-trends and brand fixed
effects. We also find evidence suggesting that response to the termination of the false claims
is heterogenous: our preliminary findings indicate that it is the new-comers that are most
impacted by these false claims and persist in their purchases even after the false claims have
been removed.
This effect is an aggregate of consumer responses to the termination of the claims as well
as possible strategic firm-side responses in terms of price and advertisement changes. The
firm-side data provide some evidence that competitors as well as the parent brand of the
impacted product respond with price declines in certain DMAs.
24
References
[1] Bronnenberg, B.J., J.P. Dube, M. Gentzkow and J.M. Shapiro (2013), “Do Pharmacists
Buy Bayer? Sophisticated Shoppers and the Brand Premium”, working paper, Chicago
Booth
[2] Burke,R.R., W. S. DeSarbo, R. L. Oliver and T. S. Robertson (1988), “Deception by
Implication: An Experimental Investigation”, Journal of Consumer Research, 14(4),
483-494
[3] Darke, P.R and R.J.B. Ritchie (2007), “The Defensive Consumer: Advertising Deception, Defensive Processing, and Distrust”, Journal of Marketing Research, 44, 114-127
[4] Dyer R.F and P.G. Kuehl (1978), “A Longitudinal Study of Corrective Advertising”,
Journal of Marketing Research, 15, 39-48
[5] Ippolito, P.M and A.D. Mathios (1990), “Information, Advertising and Health Choices:
A Study of the Cereal Market”, The RAND Journal of Economics, 21(3), 459-480
[6]
(1995), “Information and Advertising: The Case of Fat Consumption in the United
States”, The American Economic Review, 85(2), 91-95
[7] Jin, G. and A. Kato (2006), “Price, Quality, and Reputation: Evidence from an Online
Field Experiment.” The RAND Journal of Economics, 37(4), 983-1005
[8] Jin, G. and P. Leslie (2003), “The Effects of Information on Product Quality: Evidence
from Restaurant Hygiene Grade Cards”, Quarterly Journal of Economics, 118(2), 409-51
[9] Johar, G.V. (1995), “Consumer Involvement and Deception from Implied Advertising
Claims”, Journal of Marketing Research, 32(3), 267-279
[10] Moorman, C. (1996), “A Quasi Experiment to Assess the Consumer and Informational
Determinants of Nutrition Information Processing Activities: The Case of the Nutrition
Labeling and Education Act”, Journal of Public Policy & Management, 15 (1), 28-44
[11] Moorman, C., R. Ferraro and J. Huber (2012), “Unintended Nutrition Consequences:
Firm Responses to the Nutrition Labeling and Education Act”, Marketing Science,
31(5), 717-737
[12] Nelson, P. (1970), “Information and Consumer Behavior”, Journal of Political Economy,
Vol. 78 (2), 311-329
25
[13] Roe,B., A. S. Levy and B. M. Derby (1999), “The Impact of Health Claims on Consumer
Search and Product Evaluation Outcomes: Results from FDA Experimental Data”,
Journal of Public Policy & Marketing, 18(1), 89-105
[14] Peltzman, S. (1981), “The Effects of FTC Advertising Regulation”, Journal of Law and
Economics, 24 (3), 403-48
[15] Rossi, P.E, R. E. McCulloch and G. M. Allenby (1996): “The Value of Purchase History
Data in Target Marketing”, Marketing Science, 15 (4), 321-340
[16] Stigler, L. (1961), “The Economics of Information”, Journal of Political Economy, Vol.
69 (3), 213-225
[17] Zinman, J. and Zitzewitz, Z. (2013), “Wintertime for Deceptive Advertising”, working
paper, Dartmouth College
26
A
Demand Estimates
Table 14: Demand Estimates: RTE Cereal
S1
Kel Frosted Mini-Wheats
False Claims
FTC
G M Cheerios
False Claims
S2
0.033
0.063
(32.50) (48.55)
-0.013
-0.012
(-46.46) (-31.75)
-0.009
-0.090
(-6.76) (-38.24)
-0.003
0.026
(-9.16) (56.55)
FTC
G M Honey Nut Cheerios
False Claims
FTC
Post Honey Bunches of Oats
False Claims
FTC
Kel Frosted Flakes
False Claims
0.006
0.037
(4.07) (17.44)
-0.001
0.061
(-3.86) (106.26)
-0.018
0.064
(-13.89) (38.98)
-0.012
-0.005
(-33.78) (-10.95)
0.016
0.077
(7.89) (28.63)
0.004
0.092
(8.75) (145.73)
FTC
G M Cinnamon Toast Crunch
False Claims
FTC
Kel Rice Krispies
False Claims
FTC
Kel Raisin Bran
False Claims
FTC
CTL BR Frosted Mini-Wheats
False Claims
0.000
(-0.00)
-0.003
(-6.09)
0.016
(6.04)
0.046
(62.63)
0.011
(5.09)
-0.009
(-20.44)
-0.021
(-7.02)
0.016
(21.93)
0.018
(8.01)
-0.005
(-10.31)
0.049
(18.94)
0.014
(25.24)
0.003
0.019
(1.25)
(7.68)
FTC
0.005
0.021
(8.17) (33.82)
Log Likelihood
-923820 -627266
N hh
5,000
5,000
N obs
725,866 725,866
Brand constants (Before, During, After)
Yes
Yes
Price
Yes
27
Table 15: Demand Estimates: Yogurt
S1
Dannon Activia
False Claims
S2
0.024
0.004
(188.11) (26.78)
-0.013
0.038
(-42.85) (107.90)
FTC
Yoplait
False Claims
FTC
Stonyfield
False Claims
FTC
Dannon
False Claims
0.009
(188.11)
0.001
(6.08)
0.004
(26.78)
0.015
(69.97)
0.009
(177.11)
-0.011
(-15.92)
0.004
(54.28)
0.051
(76.15)
0.015
0.025
(188.50) (207.28)
0.001
0.030
(5.46) (99.86)
FTC
Yoplait Whips!
False Claims
0.007
(64.81)
0.000
(0.07)
FTC
Chobani
False Claims
0.002
(20.43)
0.000
(0.84)
0.111
0.078
(314.54) (218.20)
0.024
0.031
(114.00) (139.81)
FTC
Log Likelihood
-1304160 -1090260
N hh
5,000
5,000
N obs
892,061 892,061
Brand constants (During, After)
Yes
Yes
Price
Yes
28
Table 16: Demand Estimates: Yogurt Drinks
S1
Dannon Dan Active
False Claims
S2
0.032
0.035
(358.41) (391.74)
-0.040
-0.026
(-73.14) (-46.21)
FTC
Dannon Danimals
False Claims
0.042
0.039
(153.78) (142.33)
-0.022
-0.016
(-19.21) (-13.63)
FTC
Lifeway
False Claims
0.053
0.055
(146.57) (150.28)
0.010
0.017
(7.44) (13.57)
FTC
Stonyfield
False Claims
-0.014
-0.015
(-38.98) (-42.27)
-0.018
-0.002
(-5.26) (-0.71)
FTC
Log Likelihood
-176867 -175890
N hh
8,828
8,828
N obs
166,080 166,080
Brand constants (During, After)
Yes
Yes
Price
Yes
29
Table 17: Demand Estimates: Nutritional Supplements
S1
S2
Airborne
False Claims
-1.038
0.109
(-179.19) (17.70)
-1.682
-0.604
(-366.49) (-126.75)
FTC
Nature Made
False Claims
-2.410
-2.155
(-209.48) (-186.64)
-2.446
-2.168
(-311.15) (-275.09)
FTC
Nature’s Bounty
False Claims
-2.903
-2.658
(-203.00) (-185.04)
-2.581
-2.343
(-339.25) (-307.51)
FTC
Emergen-C
False Claims
-3.648
-2.143
(-181.42) (-104.54)
-3.470
-1.416
(-287.09) (-115.61)
FTC
Log Likelihood
N hh
N obs
Brand-specific No-Flu season constants (During, After)
Price
30
-295163
10,798
348,359
Yes
-292768
10,798
348,359
Yes
Yes