bundles. The time-watching data allows the authors to attribute greater value to those
channels that are viewed more intensely.
In
Crawford and Yurukoglu (2012), the utility of household i from time t
ic
spent
watching channel c in bundle j is
ν
ij
¼
X
c2C
j
γ
ic
log 1 + t
ic
ðÞ, (3.11)
where C
j
is the set of channels in cable bundle j.
4
The consumer allocates total hours per
day across the various channels plus an outside good of not watching television. The taste
parameters γ
ic
depend on consumer demographics. Consumer utility for bundle j then
depends on the optimized values of the channel utility as well as on the price of the bun-
dle, on the observed (z
j
) and unobserved (ξ
j
) characteristics of the bundle, and on random
“logit” error terms
u
ij
¼ν
ij
+ z
j
ψ αp
j
+ ξ
j
+ E
ij
: (3.12)
Estimation proceeds as in BLP. The unobserved characteristics include important features
such as the quality of bundled Internet service. Therefore, as usual, price and the unob-
served characteristics are likely highly correlated. The instruments for price are direct
“Hausman”-style instruments. These are the prices of similar cable bundles as charged
by other firms in the same DMA. These instruments are correlated with price if there
are common within-DMA shocks, as seems very likely to be true. They are uncorrelated
with ξ
j
if unobserved product characteristics (and other firm/bundle demand “shocks”)
are independent within DMAs. As usual with Hausman instruments, this second require-
ment could be criticized. However, the variation in other-firm price is a very powerful
instrument if correctly specified.
3.3. ADVERTISER DEMAND
Audience demand is an essential component of models of revenue, but the translation of
audience demand into revenue requires models of advertiser demand.
The simplest approach is to assume that the medium in question is small relative to the
advertising market, in which case the media firm is a price taker in the advertising market.
In such a case, the price per media consumer might nevertheless vary by type of con-
sumer, as consumers vary in their value to advertisers depending on age, gender, income,
4
Crawford et al. (2015) extend this Cobb–Douglas utility function to a more flexible form that allows for
different degrees of diminishing marginal utility across channel types. In particular, sports channels appear
to be a very valuable part of cable bundles even though time spent watching is not extraordinarily high.
One possible explanation is that the first hours spent watching sports have a very high value, but there is also
a relatively steep decline in marginal utility (e.g., once the highly valued top sports events have been
viewed, extra hours of watching other events give little marginal utility).
105
Empirical Modeling for Economics of the Media
and so forth. In some cases, advertisers may care about the racial or ethnic composition of
the audience, perhaps because ethnicity is correlated with preferences for different adver-
tised goods.
It might also be true that advertiser demand is downward sloping—that is, that the
marginal willingness to purchase an advertisement is declining in the size of the total audi-
ence exposed to the ad, or in the total demographics (quantity times demographic share)
of the audience. More subtly, advertisers may care about the number of times a given
person is exposed to an ad, with a willingness to pay that declines in the degree of repeat
exposures to the same person.
Berry and Waldfogel (1999) observe the market-level price of radio advertising as rev-
enue per radio listener. They model advertiser demand as a simple constant-elasticity
function of total market listening share, as in
ln r
t
ðÞ¼x
t
λ
1
λ
2
ln S
t
ðÞ+ ν
t
, (3.13)
where x
t
is a vector of market characteristics and S
t
is the total share listening to radio. The
unobservable advertising demand shock ν
t
is likely correlated with the listening share,
since a high level of ad demand will encourage station entry, which in turn leads to high
listening. Possible instruments for listening share include the same variables discussed
above in the context of nested logit radio demand. These include population and pop-
ulation by demographics.
With richer data, the specification in
(3.13) can be extended. Sweeting (2013) models
station-level ad prices and includes controls for station format and audience demo-
graphics. Jezkiorski, among others, attempts to model prices at the level of a 60-second
ad slot. This introduces the “quantity of advertising,” a
t
, in addition to the quantity of
listeners. In radio, the intensity of advertising presumably drives down listening share
so that demand for listening should depend on a possibly endogenous quantity of ads.
It will also affect the supply side, discussed below.
Rysman (2004) considers advertiser demand for traditional printed “yellow pages”
directories that consist of paid advertisements for various, mostly consumer-oriented,
retail and service establishments. Pre-Internet, these were a primary reference for con-
sumers seeking service providers and retail outlets. The two sides of the market are
the advertisers, who generate revenue for the directory, and consumer “users” who
receive the yellow pages book for free. Advertiser demand presumably increases in
the amount of usage and usage presumably increases in the amount of advertising (which
increases the information value of the directory). This creates a classic network effect
among competing oligopoly firms. Rysman’s model features consumers who “single-
home” each given search in a single yellow book and features advertisers whose
“profit per look” is constant (so that competing directories are neither complements
nor substitutes in price). In particular, he derives a Cobb–Douglas directory-level aggre-
gate inverse advertising demand function for directory j in market t of
106 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.