Deceptive Claims using Fake News Marketing: The Impact on Consumers Anita Rao Booth School of Business University of Chicago February 12, 2018 THIS IS A DRAFT. COMMENTS ARE WELCOME, BUT PLEASE DO NOT CITE WITHOUT AUTHOR’S PERMISSION Abstract Fake news can be harmful if it misleads consumers to take actions they otherwise would not have taken (e.g. vote for another party, purchase an inferior product). How- ever, if fake news merely confirms existing beliefs without changing consumers’ actions, the extent of such harm is less severe. The main challenge in identifying the impact of fake news, is that we do not observe actions before and after the exposure to fake news. This paper exploits a unique setting where the FTC enabled the shutdown of ten companies that were operating fake news websites that in reality were advertise- ments for various products. Using detailed browsing data of these product websites, I identify the extent of consumer interest in the presence and absence of fake news. The findings indicate that interest wanes after the shutdown of fake news, but there is some substitution to other channels such as regular advertisements. 1
20
Embed
Deceptive Claims using Fake News Marketing: The Impact on ... · Fake news can be harmful if it misleads consumers to take actions they otherwise would not have taken (e.g. vote for
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Deceptive Claims using Fake News Marketing: The
Impact on Consumers
Anita Rao
Booth School of Business
University of Chicago
February 12, 2018
THIS IS A DRAFT. COMMENTS ARE WELCOME, BUT PLEASE DO NOT CITE
WITHOUT AUTHOR’S PERMISSION
Abstract
Fake news can be harmful if it misleads consumers to take actions they otherwise
would not have taken (e.g. vote for another party, purchase an inferior product). How-
ever, if fake news merely confirms existing beliefs without changing consumers’ actions,
the extent of such harm is less severe. The main challenge in identifying the impact
of fake news, is that we do not observe actions before and after the exposure to fake
news. This paper exploits a unique setting where the FTC enabled the shutdown of
ten companies that were operating fake news websites that in reality were advertise-
ments for various products. Using detailed browsing data of these product websites, I
identify the extent of consumer interest in the presence and absence of fake news. The
findings indicate that interest wanes after the shutdown of fake news, but there is some
substitution to other channels such as regular advertisements.
1
1 Introduction
Fake news sites have recently garnered a lot of attention because of their potential impact
on the 2016 elections. Sites such as Facebook and Google are under fire for their role as
distributors of fake news. However, the empirical impact of fake news on actual outcomes
is unknown. Alcott and Gentzkow (2017), in the context of the 2016 elections, point out
that it is possible fake news merely strengthened voters’ predetermined beliefs, but did not
change their voting behavior. For example, those likely to vote for Trump are the ones who
believe pro-Trump fake news. It is empirically challenging to measure the true impact of the
fake news, because we do not observe voting behavior prior to the exposure to fake news.
In this paper, using a set of discontinued fake news sites, I disentangle the impact of fake
news from that of regular advertising on consumers’ purchase propensity. In April 2011, the
FTC brought to halt the operations of ten fake news website operating companies. These
ten companies operated over 150 fake news websites with names such as onlinenews6.com
and consumerdigestweekly.com. The product domains that these fake news sites referred
consumers to, typically sold purported weight-loss products and colon cleansing products.
These companies used both direct advertising through sites such as google, and fake news
“ads” to propagate their products. The fake news sites falsely reported, in a journalistic
manner, the positive impact of using the products. The FTCs actions were directed toward
the fake news operating companies, and not the advertising companies, which were under no
restraint and could continue to operate. The difference in product domain visitations before
and after the FTC complaint identify the causal impact of the visitations due to the fake
news.
Using detailed browsing data from comScore, which also tracks referral domains, one
can identify whether consumers visited the product’s website directly, were referred from a
general search site such as google, or were referred from a fake news site. We also observe the
chronological sequence of website visitations, i.e., did a user first reach the product domain
via an advertisement and subsequently visited the domain directly.
The findings indicate visits to product domains that used fake news as an advertising
means drop after the shutdown of the fake news operating sites. Breaking down the visits
into organic visits or visits through advertisement referrals suggests the drop in visits comes
from a drop in referrals via fake news (mechanistically) as well as a drop in direct visits.
However, there appears to be an increase in visitations via regular advertisements, suggesting
that fake news and advertisements are substitutes atleast in these product domains.
2
Contribution The impact of fake news has largely been studied in the political domain
(e.g., Guess, Nyhan and Reifler 2018; Alcott and Gentzkow 2017), with little work in the
domain of consumer products. That consumers might choose to consume news geared to-
wards their preferences has been studied in Gentzkow and Shapiro (2006, 2010), Zhu and
Dukes (2015) and Simonov (2018) to name a few. This paper contributes to the literature on
media slant by focusing on fake news as a means of advertising, and its impact on consumer
purchases as opposed to media consumption in a political setting.
This paper also contributes to the literature on deceptive practices undertaken by firms
to win consumers’ walletshare. Rao and Wang (2017) study the impact of false health claims
on consumer demand, Zinman and Zitzewitz (2013) study ski resorts’ deceptive reporting
of snowfall to attract demand, Luca and Zervas (2016) study review fraud by restaurants,
and Chiou and Tucker (2016) study ads that present selective information. However, little
empirical work documents the impact of deceptive advertising in the form of fake news.
This paper studies advertisements geared to appear like news, bringing together the
literature that studies news consumption and literature that studies firms’ deceptive practices
and their impact on consumers purchase decisions. Unlike the previous literature this paper
studies not only what is claimed but also how it is depicted, i.e., false claims made to appear
like articles by legit news organizations. A closely related paper is Chiou and Tucker (2018)
who show that after Facebook banned fake news ads from their ad networks, sharing of
fake news articles related to anti-vaccines dropped dramatically. In this paper, I further
study what happens to consumer behavior after the ban on fake news style marketing: do
consumers nonetheless find the product domain in the absence of the fake news ad either
directly or through regular online ads?
2 Data and Descriptives
In April 2011, the FTC identified ten companies as operating fake news websites1. Table 1
lists a subset of the fake news websites operated by the ten companies as recovered from the
FTC complaint files accessed using the Bloomberg Law database. Figure 3 in the Appendix
shows an article from one of these websites, extracted from one of the FTC court dockets
(FTC v Beony International, Attachment C), which highlights the nature of the fake news.
In short, in these fake articles, the “journalists” voice their skepticism of the products, claim
to try it for themselves, and report their “fake” findings of weight loss. Figure 5 shows
another such article from another fake news operating website as cited in one of the FTC
Note: (All) refers to all domains that were referred to by the fake news sites, (Cited by FTC) restrictsattention to those sites cited in the FTC court documents, (Active) refers to those that are definitely activeafter the FTC shutdown order in the comScore data.
7
Note: Red line indicates the date of the FTC shutdown order
Figure 2: Fake News domains’ visit counts
3 Empirical Evidence
Using the FTC order date as an event, I run a before-after regression on all product domains
that were referred to by the fake news sites. In this regression, I also compare this estimate
to a placebo date. Finally, I plan to add a control group of product domains which will
provide a differences in differences estimate.
yijt = βpre1 × PreFTCt + βpost
1 × PostFTCt
+ βpre2 × PreCtrlt + βpost
2 × PostCtrlt
+ αY + αj + αi + εijt
(1)
Here yijt is the number of visits individual user i made to product domain j in month
t.PreFTC and PostFTC represent the 3-months before and after the FTC shutdown,
PreCtrl and PostCtrl represent the 3-months before and after a placebo date which was
chosen to be 1 year before the FTC shutdown order. αY is the year fixed-effect that allows
for any yearly trends, αj is the domain fixed-effect that allows for the fact that some domains
have more visits than others, αi is the individual fixed effect.
An observation in this regression is an individual-domain-year-month. The comScore
dataset records only sessions of active browsing. To allow for no-visits, I expand the dataset
8
so that every individual has an observation for every month of the data for every domain
ever visited, accounting for zero visit counts when there was no visitation to domains. This
expansion allows the individual’s search behavior to change, i.e., she might visit fewer do-
mains after the shutdown. She might also stop visiting weight-loss related websites causing
the market for such sites to shrink. Not accounting for the “zero visits” would result in an
incorrect estimate. Some sites that are referred from the fake news domains include “nor-
mal” sites such as accuweather.com and live.com and a few legitimate ad networks such as
crwdcntrl.net. To measure the impact on only the fake product domains, I eliminate such
sites2 for the regression analysis specified in Equation 1.
Table 4 presents the results of this regression. Compared to the PreFTC period, the
product domains see 0.113 fewer visits after the fake news sites were shutdown. To ensure
this is not a time-trend, the same difference for the Placebo year is 0.032 site visits. The
difference-in-difference,(βpost1 − βpre
1
)−(βpost2 − βpre
2
), is 0.081 site visits and is a statistically
significant drop. To provide further context, the average number of site visits per individual-
domain-year-month is 0.13, making this drop an economically meaningful one.
Table 4: Difference in Domain Visits after Fake News Sites were Shutdown
coeff t-stat
Post FTC βpost1 0.024 18.05
Pre FTC βpre1 0.136 30.43
Difference, FTC X (Post-Pre) βpost1 − βpre
1 -0.113 -24.11
Post Placebo βpost2 -0.110 -16.74
Pre Placebo βpre2 -0.079 -10.74
Difference, Placebo X (Post-Pre) βpost2 − βpre
2 -0.032 -3.21
Difference-in-difference estimate -0.081 -7.42N obs 2,811,129N id 55,745Fixed effects Year
DomainIndividual
Cluster Individual
2In two robustness checks, I include 1) all legitimate sites and 2) only the legitimate ad networks in theregression and find the results hold. This inclusion is done to allow for the scenario where the productdomains use legitimate sites/ad networks to advertise their products, which is possible.
9
3.1 Mechanism
The above results show that visitations to product domain sites dropped after the FTCs
efforts to halt these fake news operations. While visitations to the domains from the fake
news sites have to go to zero (by definition, since the fake news sites do not exist anymore),
it is not clear whether direct visits to these sites go up or down post-shutdown. Similarly, we
do not know whether referrals coming in through other ads/sources such as facebook go up
or down. Knowing the total impact is crucial, because if consumers find their way to these
product domains through other means, then the impact of the policy (i.e., the shutdown) is
unclear.
To this end, I classify visits as “Referred via Fake News”, “Referred via Other Sources”,
and “Direct”. The total visits to a product domain consists of one of these three forms of
visits. Table 5 presents the regression results by this classification. As expected, referrals
via fake news sites (mechanistically) drop. Direct visits drop after the fake news shutdown,
suggesting there is a treatment effect of shutting down the fake news sites. Moreover, the
magnitude of the drop is almost similar to the total drop in site visits, implying that almost
all the decline is coming from a drop in direct visits to the product domains. Interestingly,
referrals from other sites (regular ads) do not change and continue to remain the same as in
the placebo year. This could imply that there is little spillover between the fake news site
shutdown and regular ad traffic.
10
Table 5: Mechanism: Which source changed after fake news shutdown
(1) (2) (3)
Total Direct Referred via Referred viaFake News Other
Accounting for the possibility that some Regular Ads were also shutdown
It is possible that some of the “regular ads” were also shutdown by the FTC during the
investigation period. As an example, the FTC issued a civil investigative demand (CID) to
Facebook asking for documents relevant to the fake news site“new6reports.com”. Facebook’s
declaration stated that the advertiser associated with this domain had initiated other ad
campaigns, and after the receipt of the Civil Investigative Demand, Facebook had disabled
these ads. If regular advertisements were also shutdown, then the drop estimated in Tables
5 and 6 would be an overestimate because the drop occurs mechanistically via the shutdown
of the regular ads. I therefore exclude such advertisers from the analysis, and conduct a
robustness check. To do so, I classify as Regular Ads only those that remained active post-
April 2011, i.e., after the shutdown order.
Table 7 reports the estimates of the drop in visits occurring via regular ad referrals
after the shutdown order. The results in this table are consistent with the previous results,
referrals via regular ads either remain unchanged or increase after the shutdown, providing
evidence of a selection effect.
12
Table 7: Changes in referrals via regular ads
All visits First visits
coeff t-stat coeff t-stat
Post FTC βpost1 0.005 8.61 0.002 11.36
Pre FTC βpre1 0.023 10.77 0.004 25.04
Difference, FTC X Post-Pre βpost1 − βpre
1 -0.019 -8.43 -0.003 -12.08
Post Placebo βpost2 -0.073 -15.70 -0.001 -2.94
Pre Placebo βpre2 -0.049 -9.82 0.017 59.21
Difference, Placebo X Post-Pre βpost2 − βpre
2 -0.023 -3.43 -0.018 -52.45
Difference-in-difference estimateFTC-Placebo X Post-Pre 0.004 0.62 0.015 36.76
N obs 2,811,129 2,810,441N id 55,745 55,745
Fixed effects YearDomainIndividual
Cluster Individual
Note: Dependant variable is the visits to product domains arising from referrals viaregular advertisements.
Accounting for the possibility that merchants’ sites were also shutdown
Here, I check for the possibility that the merchants’ sites, i.e., domains actually selling the
products in question were shut down (either by the FTC, or pre-empted by the shut down
of their affiliate marketers). In order to do so, I first keep only the domains that are cited
in the FTC court dockets. Further, I restrict attention to those sites that have at least
one visit in the comScore data in the post-shutdown period. Table 8 presents the results
of this analysis, indicating that the drop in site visits exists even for those sites that were
active after the shutdown order. The difference in site visits in the months following the
shutdown, relative to the months before is -0.080 site visits. The same difference in the
placebo year is 0.015 visits, indicating a decline in site visits of the order of 0.095 site visits.
These number although small are economically meaningful, because they are evaluated at
the individual-domain-month level, where the average visit is 0.05 site visits.
13
Table 8: Robustness Check: Keeping only those merchants that were cited by FTC,and were active post-shutdown
Cited Activeby FTC
coeff t-stat coeff t-stat
Post FTC βpost1 -0.001 -0.41 0.007 1.16
Pre FTC βpre1 0.080 27.01 0.087 18.46
Difference, FTC X Post-Pre βpost1 − βpre
1 -0.081 -21.55 -0.080 -10.30
Post Placebo βpost2 -0.035 -14.99 0.019 4.99
Pre Placebo βpre2 -0.009 -3.00 0.004 1.07
Difference, Placebo X Post-Pre βpost2 − βpre
2 -0.026 -6.74 0.015 2.75
Difference-in-difference estimateFTC-Placebo X Post-Pre -0.055 -10.24 -0.095 -10.02N obs 234,496 87,184N id 7,903 3,418Fixed effects Year
DomainIndividual
Cluster Individual
Note: Dependant variable is the total visits to product domains. (Cited by FTC) restricts attention tothose sites cited in the FTC court documents, (Active) refers to those that are definitely active afterthe FTC shutdown order in the comScore data.
Accounting for PR effect of the FTC press-release
The decline in direct visits to the product domains could be caused by negative publicity asso-
ciated with the FTC press-release, which was picked up by major news outlets. If consumers
become skeptical of the product domains that used these fake news style advertisements,
they are less likely to visit the domains. In other words, the reason for the observed em-
pirical decline is confounded between stoppage of the fake news ads and negative publicity.
To rule out negative publicity, I focus on the 3 months prior to the FTC press-release. The
FTC conducts an investigation before the public announcement, and sites have the option
of shutting down during this investigation period.
Four sites that appear to have shutdown prior to the FTC press-release are consumertips-
daily6.com, health8news.com and online6health.com. These sites had a substantial number
of visits in January 2011 and February 2011, but no visits March 2011 onward. Therefore,
it is likely these sites were inoperative following March 2011. Looking at the product do-
mains these sites referred consumers to, in March vs. February/January 2011 will give us
14
the impact of the shutdown of this fake news website. More importantly, since the fake news
site shutdown prior to April 2011, any effect is not confounded with the PR effect of the
press-release which occurred more than a month later in April 2011. Table 9 presents the
results of this regression.
Table 9: Robustness Check: Accounting for PR effect of FTC press release
coeff t-stat
Post FTC βpost1 0.387 8.53
Pre FTC βpre1 0.772 9.63
Difference, FTC X Post-Pre βpost1 − βpre
1 -0.385 -4.17
Post Placebo βpost2 -0.884 -14.64
Pre Placebo βpre2 -0.892 -14.78
Difference, Placebo X Post-Pre βpost2 − βpre
2 0.008 0.09
Difference-in-difference estimate -0.393 -3.13N obs 149,209N id 5,697Fixed effects Year
DomainIndividual
Cluster Individual
4 Conclusion
This paper examines the role of fake news as advertisements, and finds that fake news can
cause increased interest in product domains. The shutdown of fake news seems to have
diverted some traffic to regular advertisements indicating that the two are substitutes at
least to some extent. Direct visits to the product domains using fake news websites dropped
significantly. While the current analysis cannot separately identify if the drop comes directly
from the absence of fake news or from any bad press surrounding these products during the
FTC investigation, this paper is one step toward understanding the impact of fake news on
consumers’ purchase decisions.
15
References
[1] Allcott, A. and M. Gentzkow (2017), “Social Media and Fake News in the 2016 Election”,
Journal of Economic Perspectives, 31(2), 211-236.
[2] Chiou, L. and C. Tucker (2016), “How Do Restrictions on Advertising Affect Consumer
Search?”, Available at SSRN: http://ssrn.com/abstract=1542934.
[3] Chiou, L. and C. Tucker (2018), “Fake News and Advertising on Social
Media: A Study of the Anti-Vaccination Movement”, Available at SSRN:
https://ssrn.com/abstract=3209929.
[4] Gentzkow, M. and J. M. Shapiro (2006), “Media Bias and Reputation”, Journal of
Political Economy, 114 (2): 280– 316.
[5] Gentzkow, M. and J. M. Shapiro (2010), “What drives media slant? evidence from us
daily newspapers”, Econometrica 78 (1), 35-71.
[6] Guess, A., B. Nyhan, and J. Reifler (2018), “Selective Exposure to Misinformation: Ev-
idence from the consumption of fake news during the 2016 U.S. presidential campaign”,
working paper, Dartmouth College.
[7] Luca, M. and G. Zervas (2016), “Fake It Till You Make It: Reputation, Competition,
and Yelp Review Fraud”, Management Science, 62(12), 3412-3427.
[8] Rao, A. and E. Wang (2017), “Demand for “Healthy” Products: False Claims and FTC
Regulation”, Journal of Marketing Research, 54(6), 968-989.
[9] Simonov, A. (2018), “What Drives Demand for Government-Controlled News in Rus-
sia?”, working paper, Columbia Business School.
[10] Zhu, Y. and A. Duke (2015), “Selective Reporting of Factual Content by Commercial
Media”, Journal of Marketing Research, 52, 56-76.
[11] Zinman, J. and Zitzewitz, Z. (2013), “Wintertime for Deceptive Advertising”, working
paper, Dartmouth College
16
A Fake News Article Example
Source: FTC Document, Declaration of Douglas McKenneyAccessed from: FTC v. Beony International LLC, Attachment C, Bloomberg Law
Figure 3: Example of a Fake News Article from consumertipsdaily6.com (Page 1)
17
Source: FTC Document, Declaration of Douglas McKenneyAccessed from: FTC v. Beony International LLC, Attachment C, Bloomberg Law
Figure 4: Example of a Fake News Article from consumertipsdaily6.com (Page 2)
18
Source: FTC Document, Declaration of Loretta Kraus, Coulomb Media, Inc., et al.,https://www.ftc.gov/sites/default/files/documents/cases/2011/04/110419coulomb-kraus-pt1.pdf
Figure 5: Example of a Fake News Article from consumerhealthwarning.com (Page 1)
19
B How do consumers reach fake news sites
Source: FTC v Circa Direct court document, Exhibit 5Investigator’s search for commonalities, in this case “this stuff truley”, resulted in many fake news sitesappearing organically.
Figure 6: Organic links to fake news sites
Source: FTC v Ambervine court document, McKenney Attachment BInvestigator’s search for “acai berry” resulted in sponsored ads for fake news sites
Figure 7: Sponsored Ads linking to fake news sites