consumer equals ϑ 1+λðÞ, with λ 2 0, 1½. A possible interpretation is that the probability
with which a consumer becomes aware of the ad is x on each platform. An ad to the same
consumer on another platform raises the chance that she becomes aware of the ad by
x(1 x). Therefore, normalizing the value of informing a consumer to 1, we have that
ϑ ¼x and λ ¼1 x.
In
Anderson et al.’s (2014) model, first, consumers form an expectation about the
advertising intensity on each platform. Then, platforms set a price per ad, and advertisers
rationally anticipate the number of consumers and choose to buy ads. Finally, consumers
decide which platform to join. Due to the game structure, consumers do not react to the
actual number of ads, which simplifies the solution of the game.
29
In equilibrium, each platform charges a price of (omitting arguments)
D
i
ϑ + D
12
ϑλ;
and all advertisers are active on both platforms. Hence, platforms can extract the full sur-
plus an advertiser obtains when informing an exclusive consumer, but only part of the
surplus that an advertiser obtains when informing an overlapping consumer. An advertiser
obtains a positive profit of D
12
ϑ(1 λ). Therefore, the principle of incremental pricing
reemerges.
Anderson et al. (2014) then focus on several important aspects of media markets that
are influenced by the presence of multi-homing consumers. Consider the well-known
problem of content duplication and suppose that each media platform has the choice of
providing content A or B. If consumers only single-home and more than two-thirds of
them are interested in content A instead of content B, then both platforms will specialize
in content A. This is because if half of the consumers choose platform 1 and the other half
platform 2 when both have the same content, then each platform gets more than one-
third of the consumers. By contrast, when choosing content B, a platform gets less than
one-third of the consumers. In general, duplication of content occurs in equilibrium if
and only if
D
A
2
> D
B
;
where D
j
, j ¼A, B, denotes the consumership for content j.
29
The game involves passive expectations as in the classic model by Katz and Shapiro (1985). This form of
expectation building implies that consumers form their expectations before observing the prices set by
platforms.
Hagiu and Halaburda (2014) and Belleflamme and Peitz (2014) use related assumptions in
two-sided market models, in which all or a fraction of the agents of one side cannot observe the prices
charged on the other side. The opposite assumption involves agents forming expectations after observing
all prices and is (usually) denoted responsive expectations. For a discussion of the different implications of
the two assumptions on termination charges of communication networks, see
Hurkens and Lo
´
pez (2014).
480
Handbook of Media Economics
Now, suppose that consumers can multi-home. If both platforms have the same con-
tent, a fraction d
12
of consumers multi-home. If both platforms choose content A, then
each of them obtains a profit of
D
A
2
ϑ 1 d
12
ðÞ+ d
12
λðÞ:
By contrast, if one platform chooses content B, it obtains a profit of D
B
ϑ. Comparing the
two profits shows that choosing content A is preferred if and only if
D
A
2
1 d
12
1 λðÞðÞ> D
B
:
Compared to the situation without multi-homing consumers, it is evident that content
duplication occurs under a strictly smaller parameter range. Hence, if multi-homing
consumers are present, the problem of content duplication is less severe.
To sum up, the work on multi-homing provides two important lessons. First, plat-
forms set their tariffs for advertising according to incremental pricing. This is because the
consumer’s first impression is usually more valuable than the second, and consumers can
now be reached on multiple platforms. Therefore, although consumers are no longer the
bottleneck, platforms cannot extract the full surplus from advertisers. Second, platforms
do not care only about the size of demand, but also about how it is composed of single-
and multi-homing consumers; thus, the composition of demand affects market
outcomes.
10.4.2 Search Engines and Search Bias
So far, we have simplified our presentation by assuming that users know which websites’
content they are interested in and can access it directly. This implies that no intermediary
is needed to help users select their preferred websites. However, the Internet offers a mul-
titude of information, and finding the most relevant bits is often impossible without the
help of a search engine. In fact, worldwide, there are billions of queries each day on dif-
ferent search engines. As described in
Section 10.2, the most prominent one is clearly
Google, with a market share of more than 90% in European countries and a global aver-
age of more than 80%. However, other search engines dominate in some countries; for
example, in China, the search engine Baidu has a market share of more than 75%,
whereas Google China has only slightly more than 15%.
30
If the only role of search engines were to efficiently allocate users to their preferred
websites, then search engines would not be particularly relevant to this chapter. In that
limited role, the information gatekeeper directs users to the appropriate media content,
and its presence has no economic implications. However, the business model of a search
30
See http://www.chinainternetwatch.com/category/search-engine/ (accessed April 25, 2014).
481The Economics of Internet Media
engine, like that of most media platforms, centers around attracting users and obtaining
revenues from advertisers. It is therefore not obvious that the incentives of users and
search engines are perfectly aligned. In particular, search engines may bias their search
results to obtain high revenues from advertisers. Suppose, for example, that a user wants
to watch the video of a particular song and searches for it via Google. The video is
available through multiple video portals, such as YouTube, MyVideo, or Clipfish,
and Google can choose the order in which to display the search results. Since Google
owns YouTube but not the other video portals, Google may have an incentive to bias
its search results in favor of its own video portal and away from others.
Establishing such a search engine bias empirically is not always straightforward.
Tarantino (2013) reports that, in response to a query with the keyword “finance,” Google
lists Google Finance first, whereas Yahoo! lists Yahoo! Finance first. This suggests that
at least one of the two search engines is biased if consumers on one search engine are
comparable to those on the other.
Edelman and Lai (2015) consider the following quasi-experiment: In December
2011, Google introduced a tool called Google Flight Search, which helps users to search
for a flight from A to B. When Google Flight Search appeared, it always appeared in a box
at the top position. However, the appearance of Google Flight Search was very unsyste-
matic, and minor changes in the entry could lead to the appearance or disappearance of
the box.
31
Edelman and Lai (2015) estimate the change in the CTR when the Google
Flight Search box appeared. They find that with the box, the CTR for paid advertising
increased by 65%, whereas the CTRs for non-paid search of other travel agencies
decreased by 55%. Therefore, the study provides evidence that search engines are able
to influence user behavior by the layout and format with which the search results are
presented.
Search engines usually have two different kinds of links, organic (or non-paid-for)
links and sponsored (or paid-for) links. The organic links reflect the relative importance
or relevance of listings according to some algorithm. The sponsored links are paid for by
advertisers. As outlined in
Section 10.2, selling those advertising slots represents the larg-
est revenue source of Internet advertising.
The major commercial search engines sell the sponsored links via second-price auc-
tions with a reserve price for each auction. Since a search engine observes whether or not
a user clicks on a sponsored link, advertisers pay per click. Thus, the price is called the per-
click price (PCP). If an advertiser bids a higher PCP, this secures a rank closer to the top.
However, the advertiser with the highest bid does not necessarily receive the first slot on
top of page one of the search results. The search engine’s goal is to maximize revenue
from selling slots and therefore it also takes into account the number of times users click
31
For example, the box was shown when typing in “flights to orlando,” but it did not appear when searching
for “flights to orlando fl.”
482
Handbook of Media Economics
on an ad. As a consequence, the search engine needs to estimate the CTR and may put
ads with a lower PCP in a higher position if their CTR is high. Google uses a quality
score that reflects the estimated CTR to determine the slots for the respective adver-
tisers.
32
There are several studies analyzing the auction mechanism in detail, including
the seminal papers by
Edelman et al. (2007) and Varian (2007). More recent papers
are
Katona and Sarvary (2008), who analyze the interaction between sponsored and
organic links, and
B€orgers et al. (2013), who explore the bidding behavior for sponsored
links on Yahoo’s search pages.
Do search engines list search results in the best interest of consumers? The economics
literature has uncovered several reasons why search engines may have an incentive to bias
their search results. We start with reasons that are to be considered even if a search engine
is not integrated with a media platform. First, distinguishing between organic and spon-
sored links can provide one answer to why search engines bias their search results. As
Xu
et al. (2012)
, Taylor (2013a), and White (2013) point out, organic links give producers a
free substitute to sponsored links on the search engine. Therefore, if the search engine
provides high quality in its organic links, it cannibalizes its revenue from sponsored links.
At the same time, providing better (i.e., more reliable) organic search results makes the
search engine more attractive. If consumers have search costs, a more attractive search
engine obtains a larger demand. However, if the latter effect is (partially) dominated
by self-cannibalization, a search engine optimally distorts its organic search results.
Chen and He (2011) and Eliaz and Spiegler (2011) provide a further reason why
search engines may bias their search results. Since the search engine obtains profits from
advertisers, it is in its best interest that advertisers’ valuation of sponsored links is high.
This valuation increases if product market competition between advertisers is relatively
mild. Therefore, the search engine may distort search results to relax product market
competition between advertisers. In
Chen and He (2011) and Eliaz and Spiegler
(2011)
, the search engine has an incentive to decrease the relevance of its search results,
thereby discouraging users from searching extensively. This quality degradation leads to
lower competition between producers and therefore to higher prices.
33
We now turn to the case in which the search engine is integrated with a media plat-
form (as is the case with YouTube and Google). Does this lead to additional worries
about search engine bias, or can integration possibly reduce search engine bias? In what
follows, we present the models of
de Cornie
`
re and Taylor (2014) and Burguet et al.
(2014)
to systematically analyze the costs and benefits of search engine integration.
In
de Cornie
`
re and Taylor’s (2014) model, there are a monopoly search engine i ¼0
and two media platforms i ¼1, 2. The media platforms are located at the ends of a
Hotelling line, with platform 1 located at point 0 and platform 2 at point 1. Users are
32
For a more detailed discussion, see Evans (2008).
33
See Xu et al. (2010, 2011) for related models.
483
The Economics of Internet Media
distributed on the unit interval, but before deciding to search, they are not aware of their
location. This implies that without searching, a user cannot identify which media plat-
form has the content she is interested in most. A user incurs search costs s when using the
search engine, where s is distributed according to a cumulative distribution function
denoted by F.
Both the media platforms and the search engine obtain revenues exclusively from
advertising. The quantity of advertising on website i is denoted by a
i
. Users dislike adver-
tising, implying that the disutility of a user who will be directed by the search engine to
website i is γ
i
(a
i
), which is strictly increasing. A user’s utility is also decreasing in the dis-
tance between her location and the location of website i. The utility a user receives from
website i is
vd, a
i
ðÞ
¼ud
ðÞ
γ
i
a
i
ðÞ
s;
where d denotes the distance between the location of website i and the location of the
user and u
0
dðÞ< 0.
The search engine works as follows: If a user decides to use the search engine, she
enters a query. The search engine then maps the user’s query into a latent location on
the Hotelling segment and directs the user to one of the platforms. The search engine’s
decision rule is a threshold rule such that all users with x < x are directed to platform 1 and
those with x x are directed to platform 2.
Advertising is informative, and there is a representative advertiser. The expected per-
user revenue of the advertiser is
Ra
0
, a
1
, a
2
, xðÞ¼r
0
a
0
ðÞ+ xr
1
a
0
, a
1
ðÞ+1xðÞr
2
a
0
, a
2
ðÞ;
where r
i
, i ¼0, 1, 2 represents the revenue from contacting users on the search engines
or the respective media platform. A key assumption is that ads on the search engine
and on the media platforms are imperfect substitutes. That is, the marginal value of an
ad on one outlet decreases as the number of advertisements on the other outlet increases.
Formally,
@
2
r
i
a
0
, a
i
ðÞ
@a
0
@a
i
0:
This implies that the advertising revenue generated by a media platform falls if a
0
rises.
The advertiser pays platforms on a per-impression basis, and the respective prices are
denoted by p
i
. Therefore, the expected per-user profit of an advertiser is
π
a
¼Ra
0
, a
1
, a
2
, xðÞa
0
p
0
xa
1
p
1
1 xðÞa
2
p
2
:
Given that a fraction of users D use the search engine, the profit of the search engine
is π
0
¼Da
0
p
0
, while the profits of the media platforms are π
1
¼xDa
1
p
1
and
π
2
¼ 1 xðÞDa
2
p
2
, respectively. To simplify the exposition, de Cornie
`
re and Taylor
(2014)
keep a
0
fixed and focus on the choice of a
1
, a
2
, and x.
484 Handbook of Media Economics
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.