A Two-Sided Market Analysis

of Radio Ownership Caps

Catherine Tyler Mooney∗

Department of Economics
University of Oklahoma
729 Elm Ave, Norman, OK 73019
ctmooney@ou.edu

July 26, 2010

Abstract
A relaxation of radio station ownership caps in 1996 induced a wave of mergers and ac-
quisitions. This paper develops a “two-sided market” empirical model of the commercial radio
industry to analyze the effects of the change in market structure. In the model, both listener
and advertiser preferences affect one another via the advertising and programming strategies
of broadcasters, and audience demographics impact both listener and advertiser demand. The
estimation algorithm expands on that of Berry, Levinsohn, and Pakes (1995) by embedding
the computation of equilibrium advertising levels for every radio station. The estimated model
reveals modest increases in audience welfare and advertising with consolidation. Firms benefit
from increased market power by increasing advertising time in less segmented radio markets
and, likewise, decreasing advertising in more segmented markets.

Keywords: ownership structure, advertising, structural estimation, media economics, merg-

ers, random coefficients logit

∗
ACKNOWLEDGEMENTS: Financial support for this project has been provided by the American Association
of University Women and the Bankard Fund for Political Economy at the University of Virginia. The author would
like to thank Simon Anderson, Steven Stern, and Federico Ciliberto for all of their helpful guidance and comments.
Thanks to Zenobia Mehta Sribnick for help acquiring and understanding the radio station data. The author also
thanks Andrew Sweeting, Marc Rysman, Cynthia Rogers, Carlos Lamarche, Timothy Dunne, the editor, associate
editor, referees, and many seminar participants for useful comments and discussion.
1 Introduction

The 1996 Telecommunications Act substantially loosened long-standing U.S. radio ownership reg-

ulations by eliminating the forty station cap on national ownership and nearly doubling local

ownership caps. Following the policy change, the number of radio station owners in the U.S. fell

from over 2,000 to approximately 1,600, and the average local station group rose from three to

4.7 between 1997 and 2003. In large markets, some station groups grew to over ten jointly owned

stations.

Using data from the beginning and end of the wave of commercial radio mergers and acquisitions,

this paper estimates a structural econometric model of competition among broadcasters who face

the challenge of attracting demand from the audience as well as the advertisers. The substantive

research on “two-sided markets,” suggests that the predictions of standard microeconomic theory

do not always apply in this context. The theoretical foundation for this paper is Anderson and

Coate (2005)’s model of broadcasting. Firms earn revenues by including advertisements in their

broadcasts, and the advertisements create disutility for the audience. On the other side of the

market, advertisers place a higher value on larger audiences. The model presented here further

allows advertisers to have preferences over the demographic composition of the audience.

As in Berry and Waldfogel (1999), stations compete for revenue in a static game of complete

information. I expand on the previous model by allowing heterogeneous, multi-product firms to

maximize advertising revenue by choosing their stations’ ad levels conditional on entry, ownership,

and programming formats. In this model, firms may exercise market power by either increasing

or decreasing their broadcast time devoted to advertising, a feature not incorporated in other

structural empirical two-sided market analyses of media mergers, such as Argentesi and Filistrucchi

(2007) and Fan (2010). Like the other papers, also see Chandra and Collard-Wexler (2009)

and Rysman (2009), I find that post-merger outcomes differ from that of a “one-sided” market.

1
Equilibrium advertising quantities sometimes fall and often rise depending on firms’ relative market

power on each side of the market. Unfortunately, advertising levels are not observed in the data, so

some of the interesting variation, especially in station-specific advertising strategies, is not observed.

However, I am able to use the structural model to solve for advertising time using the first-order

conditions for the firm, revealing differing incentives across stations.

Another contribution is the use of a detailed specification of listener utility, allowing listeners’

preferences to depend on their demographic characteristics and including data that indicate the ex-

tent to which music overlaps among programming formats. This data proves important in correctly

describing audience substitution patterns across stations, which play a large role in identifying the

advertising equilibrium. Estimates of the firms’ revenue functions and the listeners’ utility func-

tions allow the effects of mergers on listener and firm welfare to be quantified. Another upshot of

the model is that it allows for counter-factual simulations. Using the predicted model, I isolate the

1998 to 2003 effect of station consolidation on firm behavior and welfare by controlling for changes

in population and demographics over time.

1.1 Relationship to Previous Literature

Research commissioned by the Federal Communications Commission examines the impact of the

change in ownership regulations, which are detailed in Appendix A, on the radio advertising market.

Brown and Williams (2002) find small, but positive effects of a fifty percent increase in concentration

on radio ad prices from 1996 to 2001. Chipty (2007) finds that local market concentration does

not impact market ad prices, but it does lower ratings in some markets. I also find small net effects

on ad prices, but the focus on aggregate effects is inappropriate. Outcomes vary systematically

with market conditions because the two-sided nature of the market can imply either an increase

or a decrease in advertising with concentration. I find that jointly owned stations air more ads,

knowing that lost listeners may switch to a sister station as opposed to a competitor. However,

2
especially in some very segmented radio markets, a large firm may have the ability to raise prices

by reducing advertising time.

Many radio ownership changes were accompanied by format changes. Romeo and Dick (2003)

find that many of the changes can be explained by cost-sharing in larger national radio networks, but

stations also strategically reposition their programming to increase ratings. Berry and Waldfogel

(2001) find that variety, measured by the number of formats and the number of formats per station,

increased a bit from 1993 to 1997. Sweeting (2010) confirms these results for the period from 1998

to 2001 by showing that stations in the same format category diversify their music playlists when

held by a common owner. The welfare results presented in Section 7.3.2 show small benefits to the

audience with consolidation, which is consistent with the slight increases in variety found in the

previous research, but the effects vary across demographic groups.

Political debate over radio ownership caps and other media ownership regulations has continued

for over a decade. The public good nature of the broadcasting industry and the two-sided market

model complicate standard economic analysis. Thus, policy makers and antitrust enforcement have

developed a real need for reliable evidence regarding the strategic incentives of firms in the face

of decreased competition from other radio stations. This paper fills that gap by providing a tool

for analyzing a more consolidated radio industry, as well as policy changes and mergers in other

two-sided markets.

2 Data

The data are a repeated cross-section containing variation in market size, demographics, station

characteristics, and ownership, with local caps varying over time and across markets. Over 10,000

of approximately 14,000 radio stations in the United States operate for-profit. In 2003 the average

metropolitan area had sixteen commercial stations owned by nine firms, and the ten largest firms

3
owned about 2,500 stations nationally. Data from M-Street Publications provide describe all the

commercial radio stations in the U.S., including their antennae power, FM or AM service, call sign

and frequency, programming format, and ownership. Table 1 provides summary statistics.

Most radio broadcasts cover the large core of their local metropolitan area. However, the extent

of coverage varies with antennae power and geographic features of the city. Because almost all radio

listeners move around the city during the course of the day, the industry and most researchers con-

sider the entire metropolitan area and all of the stations broadcasting within it as a single market.

To avoid the complication of multi-market strategies for stations that reach several metropolitan

areas, I restrict the data used here to 263 geographically isolated markets, those where at least

sixty percent of listening is to stations broadcasting from inside the market.

Broadcast radio is a public good to listeners. Over ninety percent of all Americans listen to

the radio at some point during the course of the average week. According to the rating service,

Arbitron (2004), during the average quarter hour of the day, about twelve percent of the population

tunes in. Quarterly hour listening reaches over twenty percent during the morning drive-time, and

drops off steeply during the evening. To capture this daily pattern, I estimate demand using listener

ratings and advertising prices for each of four day-parts: Morning (6 to 10 AM), Day (10 AM to 3

PM), Afternoon (3 to 7 PM), and Evening (7 PM to midnight).

Arbitron approximates these and all listening figures from survey data. Paid listeners record

their listening behavior in a diary for six weeks. The surveys take place in most markets semi-

annually. For the widely used AQH (average quarter hour) rating, a single individual is counted as

a listener for each five-minute period that they tune in to a station in a particular quarter hour.

The average number of counted listeners a station receives over all quarter hours in the six weeks

for each day-part divided by the total market population is the station’s rating. These figures serve

as station market shares in estimation, as in Berry and Waldfogel (1999).

4
 Table 1: Radio Station Summary Statistics for 1998 and 2003
1998 2003
Number of Stations 2,518 3,047
Number of Observations 9,816 11,866

Variable Mean Std. Dev. Mean Std. Dev.

Antenna Power (kW) 1.0 0.2 1.0 0.2
FM Service 69.0 % 46.2 % 71.7 % 45.0 %
Independently Owned 29.7 % 45.7 % 21.1 % 40.8 %
Country Format 15.6 % 36.3 % 12.0 % 32.5 %
Talk Format 15.3 % 36.0 % 16.9 % 37.5 %
Adult Contemp. Format 14.0 % 34.7 % 10.3 % 38.9 %
Rock Format 14.0 % 34.7 % 14.9 % 35.6 %
Contemp. Hits (Pop) Format 11.2 % 31.5 % 9.2 % 28.9 %
Other (Classical) Format 8.3 % 27.6 % 6.1 % 23.9 %
Religious Format 8.2 % 27.4 % 6.8 % 25.2 %
Oldies Format 7.0 % 25.5 % 8.8 % 28.3 %
Spanish Language Format 3.5 % 18.4 % 5.9 % 23.5 %
Urban Format 3.0 % 16.9 % 9.1 % 28.8 %
Years since Format or Call Sign Change 1.8 0.6 1.7 0.6
Rated Listeners per Market 7,160 8,179 6,937 8,575
· Morning 9,418 9,555 8,909 9,913
· Day 9,265 9,082 9,043 9,742
· PM 7,994 7,582 7,346 8,154
· Evening 2,563 2,653 2,450 2,694
Station Ratings 0.7 % 0.8 % 0.6 % 0.6 %
Ad Price (per person-minute)
· Morning 0.94 ¢ 0.34 ¢ 1.03 ¢ 0.37 ¢
· Day 0.78 ¢ 0.29 ¢ 0.99 ¢ 0.35 ¢
· PM 0.96 ¢ 0.36 ¢ 1.15 ¢ 0.40 ¢
· Evening 1.02 ¢ 0.38 ¢ 0.68 ¢ 0.39 ¢
NOTE: Listening and rating figures are on a quarter hourly basis.

5
 Stations differentiate themselves through their programming. I rely on the M-Street data to

group the station reported formats into the categorization used here. I add an “Other” category

to include stations of less common formats, mostly “Jazz” and “Classical” stations. From Table

1, the share of stations in some of the most popular programming formats fell a bit from 1998 to

2003, while the share of Spanish language stations and, especially, of the “Urban” format increased

substantially. The decrease in the time since the average station’s last format change reveals a

relationship between radio consolidation and format changes as noted in the previous literature.

Programming formats are not mutually exclusive categories. Many songs are played on stations

of various formats. To capture the overlap among formats, I obtain data from Mediabase 24/7,

also used in Sweeting (2010). Mediabase tracks the playlists of radio stations throughout the U.S.

I use the 2001 lists of the top 500 artists played nationally in each format to define the extent of

programming overlap among the formats. The details on constructing the format overlap matrix are

in Appendix B. I use the format overlap measures in estimating the listeners’ utility functions. The

interaction of these data with listener demographics allows for better informed listener substitution

patterns than format dummy variables.

Current Population Survey data, provided by IPUMS, describes the distribution of demograph-

ics for the potential audience (Ruggles et al., 2004). The demographic characteristics included are

age, gender, race, Hispanic origin, educational attainment, earnings, and census region. I supple-

ment this data with Census 2000 data on commuting patterns and the distribution of demographics

for smaller markets. I match each Arbitron defined radio market to the relevant Metropolitan Sta-

tistical Area.

Because radio is free to listeners, advertisers fund programming. Approximately half of all radio

advertising spots are purchased by local businesses. I use the number of retail establishments in

a metropolitan statistical area, from the Census Bureau’s County Business Patterns database, to

6
approximate the number of potential advertisers in a market.

Market advertising prices for each day-part are published by Standard Rate and Data Service

(“SRDS”). These prices capture the reported radio advertising costs of national and regional adver-

tisers who voluntarily report to SQAD, Inc. Table 1 provides summary statistics of the advertising

prices. The data presented in the table do not include a twelve percent inflation adjustment, which

I use in the econometric analysis. A downward shift in the entire ad price distribution for the

evening in 2003 creates the outlying data points.

3 Descriptive Analysis

Two regressions describe the relationship of demand with demographics, audience and advertiser

preferences, and market structure. In the first, station market shares are a function of station

characteristics interacted with market demographics as well as measures of market power and com-

petition. I reduce the over one hundred potential format-demographic interactions to twenty-four

based on Arbitron’s Radio Today, which describes the demographic profile of each format’s audi-

ence. The results in Table 3 indicate that listener preferences vary a great deal with demographics.

However, these estimates do not take into account the joint distribution of the demographic char-

acteristics, and several of the demographics, like age and gender, vary little across markets. The

structural model deals with both of these issues. The negative sign on the dummy variable for

2003 indicates falling station ratings relative to the outside good, which may be related to the

policy change. The competitive effects show that station shares fall in markets with more stations,

especially in the same format. Station group size increases shares, but to a lesser extent if the

jointly-owned stations are of the same format category. This counter-intuitive estimate suggests

the result, to be shown by the structural model, that commonly-owned stations that are close

substitutes air more ads.

7
 Table 2: Station Ratings Regression, Dependent Variable: log(share)-log(share not listening)

Parameter Estimate Std. Errora Parameter Estimate Std. Errora

Intercept -6.85 0.24 ∗ Rock∗%Age 18-24 -0.72 0.52
Antenna Power (kW) 1.17 0.05 ∗ Adult Contemp.∗%Age 18-24 -0.30 0.92
Station 2+ years oldb 0.35 0.02 ∗ Other∗%Age 18-24 -0.81 0.72
FM Frequency Band 0.53 0.02 ∗ Oldies∗%Age 18-24 -0.07 0.68
Talk Format 0.34 0.39 Rock∗%Age 25-34 -0.36 0.57
Country Format 0.88 0.24 ∗ Adult Contemp.∗%Age 25-34 -0.36 0.93
Rock Format 0.18 0.50 Other∗%Age 25-34 1.74 0.58 ∗
Spanish Language Format -0.01 0.24 Oldies∗%Age 25-34 1.09 0.59
Adult Contemporary Format 2.07 0.76 ∗ Rock∗%Age 35-64 -0.53 0.49
Religious Format -0.10 0.24 Adult Contemp.∗%Age 35-64 2.76 0.77 ∗
Contemporary Hits Format 0.52 0.24 ∗ Other∗%Age 35-64 0.40 0.48
Urban Format 1.09 0.26 ∗ Oldies∗%Age 35-64 2.03 0.50 ∗
Rock∗%Age 65+ -0.17 0.46
Morning Day-part 0.21 0.02 ∗ Adult Contemp.∗%Age 65+ 0.06 0.75

8
Daytime Day-part 0.26 0.02 ∗ Other∗%Age 65+ 2.08 0.53 ∗
Evening Day-part -1.14 0.02 ∗ Oldies∗%Age 65+ 1.41 0.55 ∗
Talk∗%Male 1.45 0.64 ∗
2003 -0.36 0.02 ∗ Rock∗%Male 1.99 0.72 ∗
Same Firm & Format Stations -0.03 0.01 ∗ Adult Contemp.∗%Male -2.68 1.15 ∗
Same Firm Stations 0.07 3.E-03 ∗ Country∗%Black -0.76 0.22 ∗
Same Format Stations -0.08 4.E-03 ∗ Religious∗%Black 3.00 0.18 ∗
Market Stations -0.03 1.E-03 ∗ Urban∗%Black 2.85 0.26 ∗
Country ∗South 0.14 0.05 ∗
R-Square 0.43 Spanish∗%Hispanic 3.00 0.16 ∗
Observations 21,682 South∗Power -0.10 0.03 ∗
Midwest∗ Power -0.02 0.03
West∗Power 0.04 0.03
NOTE: Format characteristics are defined by the extent of artist overlap among them. The dependent variable
specification implies an assumption of logit demand for listeners as in Berry (1994).
∗ indicates significance at the 95% confidence level.
a Asymptotic standard errors, b Dummy variable for same format and call sign for at least two years
 Table 3: Market/Day-part Advertising Price Regression
Dependent Variable: log(price per listener)
Parameter Estimate Std. Error
Intercept 1.76 0.29 ∗
% Age 18-24 0.35 0.29
% Age 25-34 0.55 0.30
% Age 35-64 0.04 0.27
% Age 65+ -0.20 0.36
% Male 0.65 0.39
% Black -0.19 0.17
% Hispanic 0.12 0.11
% College Graduate 0.57 0.16 ∗
% Earnings<$25K -0.18 0.21
% Earnings $25K-$65K 0.97 0.27 ∗
% Earnings>$65K -1.81 1.40
Daytime Day-part -0.10 0.03 ∗
PM Day-part 0.09 0.03 ∗
Evening Day-part -0.18 0.04 ∗
log(Retail Establishments) 0.20 0.04 ∗
log(Population) -0.41 0.05 ∗
log(Market Stations) -0.13 0.05 ∗
log(Market Formats) -0.05 0.08
HHI Concentration Index -2.E-03 9.E-04 ∗
%Black∗HHI 4.E-03 2.E-03
%Age65+∗HHI 0.01 5.E-03 ∗
2003 -0.08 0.02 ∗
R-Square 0.44
Observations 1,044
NOTE: ∗ indicates significance at the 95% confidence level.

The log price regression, presented in Table 3, reveals the importance of audience income and

competition to advertisers. It also shows that the price per listener decreases in market population,

which suggests a downward sloping demand curve. The competitive effects are once again counter-

intuitive. Price decreases in the number of stations, but it also falls in the concentration index,

unless the percent black or over age sixty-five is very large, an indication of market segmentation.

Standard economic theory cannot explain these effects, unlike a model of quantity setting firms in

a two-sided market.

The structural model allows the two-sided market theory to indicate how competition affects

9
each side of the market, marginal advertising revenue versus competition for listeners. It uses

a carefully specified model of audience behavior to describe competition via listener substitution

patterns, and advertisers’ demographic preferences are estimated more precisely when based on the

characteristics of the radio audience rather than the entire market. Furthermore, the estimated

model can be used to measure welfare and compute counter-factual outcomes, both of which are

extremely valuable in analyzing a policy change.

4 Model

Firms maximize profits by selling commercial time to advertisers, but the ads affect listener utility.

Firms have complete information about their rivals and consumers, and the outcome is a Nash

equilibrium in advertising levels. In each market there are K stations owned by J firms. Each

firm maximizes joint profits over all of its stations. It takes into account the fact that its behavior

may impact advertising prices and that its listeners may switch to one of its other stations or a

competitor.

4.1 Listeners

In each metropolitan area radio market, indexed by m, listeners, indexed by i, choose to listen to

one radio station, indexed by k, from the set of all commercial stations in the market, Km , or to

not listen during each of the four parts of the day - morning, day, afternoon, and evening, indexed

τ . Listeners have heterogeneous, horizontally differentiated preferences regarding radio station

programming. Preferences for older, higher quality stations and for stations with less advertising

and stronger antennae power are not differentiated. I expect that listeners have a net-dislike of

advertising, but this is not directly imposed. Wilbur (2008) similarly models television viewer

demand, finding a negative effect of advertising.

Let Xk be the vector of observable characteristics of radio station k. The heterogeneous pref-

10
erences for station characteristics are defined by the vector βiτ . Let Ziτ contain the listener’s

demographic characteristics and a day-part indicator, and vi is a listener specific preference for

commercial radio. Let akτ be the number of minutes of ads on station k during day-part τ , let ξkτ

be an exogenously determined measure of quality or popularity, and let ikτ be a mean zero, i.i.d.

error. The utility that listener i receives from tuning-in to station k during day-part τ is

uikτ = Xk βiτ − γakτ + ξkτ + ikτ

βiτ = β̄ + Ziτ Λ + σvi , vi ∼ N (0, 1). (1)

If a listener chooses not to listen to any commercial radio station during day-part τ , her utility is

ui0τ = ξ0τ + i0τ . The mean utility of the outside option, ξ0τ , is normalized to zero. The variance

term σ is the same for all options. It is interpreted as the standard deviation of the taste for the

inside good, commercial radio.

I assume that Ziτ , vi , and ikτ are independent and that ikτ is distributed i.i.d., Type I Extreme

Value. Thus, each individual’s probability of listening to station k in day-part τ , given by the logit

model, is
 
exp{Xk βiτ − γakτ + ξkτ }
Dikτ = P |Ziτ , vi , (2)
1 + k∈Km exp{Xk βiτ − γakτ + ξkτ }

and the market share of station k is

Z Z
Dkτ = Dikτ (Ziτ , vi )dPZ (Ziτ )dPv (vi ). (3)

The total audience size of station k for day-part τ is Dkτ M , where M is the population of the

∂Dkτ
RR
market. The effect of ads on own-station demand is ∂akτ = −γ Dikτ (1 − Dikτ )dPZ dPv , and the

∂Dlτ
RR
effect on other stations in the market is ∂akτ =γ Dikτ Dilτ dPZ dPv .

4.2 Advertisers

The assumptions for advertisers follow Rysman (2004), except for functional form. A represen-

tative, price-taking advertiser’s marginal benefit of advertising on a radio station depends on the

11
audience demographics, the station’s quantity of advertising, and the total quantity of advertising.

The advertiser draws value from sales of its product to listeners reached by its ads who ultimately

make a purchase. The advertiser’s expected marginal profitability for each minute of ad exposure to

listener i on station k during day-part τ is πikτ . The net value of radio advertising is the expected

additional profit from product sales minus the cost of placing the ads. Each station k charges pkτ

per minute of advertising during τ . The value to the advertiser is

4 X Z Z
X 
V = πikτ Dikτ M dPZ (Ziτ )dPv (vi ) − pkτ akτ , (4)
τ =1 k∈Km

where M is the market population, and akτ is the number of ad spots purchased by the repre-

sentative advertiser from k to be aired during τ . I assume that V is additively separable across

stations and day-parts, implying that ads are neither substitutes nor complements across either.

Advertisers choose the number of ads to purchase on each station during each day-part to maximize

(4). The resulting price per listener is

RR
pkτ πikτ Dikτ dPZ (Ziτ )dPv (vi )
πkτ ≡ = , (5)
Dkτ M Dkτ

where πkτ also represents the value of the average station k, day-part τ audience member.

I assume that πikτ is linear in listener demographic characteristics, market ad exposures for

listeners with the same demographic characteristics as i, and the total ad minutes aired on station

k. In concurrent theoretical work, Stuhmeier and Wenzel (2010) also use a linear price specification.

The value of each ad exposure is defined as

X
πikτ (akτ ) ≡ Ziτ ρ − α0 akτ Dikτ M − α1 akτ . (6)
k∈Km

The parameter α0 is the slope of the aggregate demand for ad exposures, and α1 captures any

advertising congestion on a single station. If negative, it could designate an increasing value to

repetitions of the same ad. I assume that the α’s do not vary by demographic group. Given (5)

12
and (6), the partial derivatives of price per listener with respect to ads are

Z Z " ! #
∂πkτ X Dikτ
= −α0 M Dikτ (1 − γakτ ) + γ anτ Dikτ Dinτ − α1 dPZ (Ziτ )dPv (vi ),
∂akτ Dkτ
n∈Km
Z Z !
∂πlτ X Dilτ
= −α0 M Dikτ (1 − γakτ ) + γ anτ Dikτ Dinτ dPZ (Ziτ )dPv (vi ), (7)
∂akτ Dlτ
n∈Km

reflecting the effect of ads on prices via downward sloping demand and audience composition.

4.3 Radio Broadcasters

Firm j’s profits are the sum over all day-parts of the advertising revenues of the stations in the set

Kj net of fixed costs, F . The profit function for firm j is

4 X
X
Πj = πkτ Dkτ M akτ − F. (8)
τ =1 k∈Kj

Firms maximize profits by choosing the number of ads to air, akτ , for all of their stations. The
∂Πj
profit maximizing first order conditions for advertising, ∂akτ = 0, imply

 
∂πkτ X ∂πlτ ∂Dkτ X ∂Dlτ
πkτ + akτ Dkτ + Dlτ alτ = −πkτ akτ − πlτ alτ (9)
∂akτ ∂akτ ∂akτ ∂akτ
l6=k∈Kj l6=k∈Kj
| {z } | {z }
marg. rev. effect listener effect

∀τ, j, k ∈ Kj . As in Anderson and Coate (2005), the left hand side is marginal revenue, and

the right-hand side reflects the marginal effect of lost listening on revenue. The summation terms

take into account joint ownership, and both effects grow in magnitude with consolidation. For

∂πlτ
the marginal revenue effect, because ∂akτ is negative, marginal revenue becomes steeper. It falls

faster when jointly owned stations provide advertisers with similar listeners and when α0 is large.

This places downward pressure on a jointly owned station’s advertising level. The right hand

∂Dlτ
side decreases with consolidation because ∂akτ is positive. The listener effect is larger when jointly

owned stations are close substitutes and when listeners receive more disutility from ads. This places

upward pressure on the number of ads. The relative magnitudes of the marginal revenue and the

13
listener effects determine whether station advertising increases or decreases due to joint ownership.

If listeners instead liked ads, there would be a positive-positive feedback loop, resulting in high

equilibrium advertising, unambiguously decreasing with monopoly power, as found in Rysman

(2004).

The prices and audience demand for any station depend on the behavior of all the stations

in the market. A Nash equilibrium in ad quantities exists with an interior solution as long as

πkτ ∂D πlτ ∂D
P
∂akτ akτ + ∂akτ alτ ≤ 0; the listener effect of joint ownership does not outweigh the
kτ lτ
l6=k∈Kj

negative own-station effect. I verify uniqueness of the estimated equilibrium ex post.

In the long run, broadcasters also choose entry, ownership, and formats. Entry and station

acquisitions are costly and highly regulated, and Sweeting (2007) shows that format switching

costs can be substantial. Thus, I assume that these decisions are fixed in the short-run, and this

model presents the last stage of a two (or more) stage game.

5 Identification

The key identification challenge is determining the effects of advertising on listener utility and

prices. I do not observe station advertising levels. However, I establish a relationship between

ad prices, ad levels, station ratings, and their derivatives using the first-order conditions for profit

maximization, Equations (9). Thomadsen (2005) and Petrin and Train (2010) also use first-

order conditions to solve for unobservable variables. Because I observe average quarterly hour

ratings and market average prices, this identification strategy reveals shifts in the distribution of

ad levels consistent with firms’ incentives, but not station specific or hour-by-hour variation in

behavior. If the ad time for each station were observed, the supply-side equations would still be

used for identification because the advertising data would be endogenous, as in Rysman (2004).

Estimation in that setting is more efficient, but the identification strategy is similar.

14
 The first-order conditions imply particular interactions of the data that shift either the marginal

revenue or the marginal listening of each station. These terms depend critically on the demographic

interactions in the random coefficient discrete choice model and the hedonic effects in the price

equation. Fortunately, the data contain a great deal of variation in market demographics which

corresponds closely with the variation in prices and market shares. Within each market, variation in

station characteristics and market shares identify the demand side parameters, while interactions

of the derivatives of demand with market structure shift the supply side, and thus identify the

advertising parameters.

Identification of the parameters in the listener utility function, other than γ, follow similar

papers using the random coefficient logit model. In the utility function, the average popularity of

each format’s programming and the average effect of other station characteristics on shares identify

β̄. The parameter matrix Λ is identified by the variation in demographics and changes in the

station choice set across markets and over time, as well as the overlap of the music playlists for

each format. Because some demographic characteristics vary little across markets, like gender,

identification of their effects on preferences depends on supplemental data describing demand by

demographic characteristic on the national scale. It is infeasible to estimate all of over 100 potential

interactions. I restrict estimation to twenty-two elements of Λ that are clearly empirically identified.

The variance of the overall taste for commercial radio, σ, is determined by variation in the share

of the outside good beyond that explained by the characteristics of radio stations.

Advertisers primarily care about audience demographics, independent of market size and com-

petitive effects. Including the predicted demographic compositions of audiences incorporates the

information available in the station market shares in the moments describing market ad prices,

identifying ρ.

Taking as given identification of the non-advertising parameters (β̄, Λ, σ, ρ), I evaluate (9) term-

15
by-term. The first expression includes the station/day-part specific price, which is identified by

the observed market/day-part price and the demographic composition of the audience. In the

partial derivative of price, α0 is identified by the correlation of audience demographics with the

audiences of the other stations in the market and the market population. The linear functional

form assumption and variation in market shares identify α1 . The first expression on the right,

the marginal effect of own-advertising on revenue via audience size, depends on the interaction of

the own-station logit derivative with the price. This identifies γ. The pattern of station ownership

drives the other two terms. The “marginal revenue effect” sums the interaction of the sister stations’

listener shares with the cross-station derivatives of price, which depend on both aggregate demand

and audience substitution patterns. The “listener effect” sums the interaction of the cross-station

logit derivatives with the sister stations’ prices. Both effects are weighted by the quantity of ads.

These terms shift the first-order conditions, providing more variation that identifies the impact of

ads, α and γ. Because ads are unobserved, either α, γ, or the mean ad time must be normalized. I

impose the industry average mean of three minutes per quarter hour, as described in Appendix C.

6 Computation and Estimation

The observed station shares and market advertising prices provide a set of moment conditions for

estimation. The moments matching the observed Arbitron ratings to the model predicted station

market shares are similar to those used by Berry, Levinsohn, and Pakes (1995) which I refer to as

“BLP.” In accordance with the previous literature, I express listener utility as the sum of the mean

utility, δkτ , a listener specific deviation from the mean, µikτ , and an extreme value error, ikτ . In

the notation of Section 4, these are

δkτ ≡ Xk β − γakτ + ξkτ and µikτ ≡ Xk Ziτ Λ + σvi . (10)

16
Hence, listener i’s utility is uikτ = δkτ + µikτ + ikτ , and i’s demand for station k is Dikτ =
δkτ +µikτ
Pe .
1+ k∈Km eδkτ +µikτ

For each evaluation of the Generalized Method of Moments objective function, the computa-

tional algorithm computes the mean listener utility for each radio station and then solves for the

Nash equilibrium advertising levels numerically. The first step uses a contraction mapping to find

the value of δkτ that matches the observed to the model predicted market share for each station

and day-part, following BLP. The second contraction mapping iterates on Equations (9) until no

firm wants to change its quantity of advertising for any of its stations for any day-part. Details are

provided in Appendix C.

In the system of first-order conditions, each station’s ad quantity depends on the ad quantities

of all the stations, including its own, which enter through the logit demand and price expressions.

Because ad time is bound between zero and fifteen minutes each quarter hour, Equations (9) are a

Km dimensional mapping from the interval [0, 15] into itself. It is continuous as long as all stations

are at an interior optimum, which is reasonable to assume for this industry. Therefore, existence

of a solution follows from Brouwer’s Fixed Point Theorem. A sufficient condition for uniqueness
∂ 2 Πj ∂2Π
≤ | ∂a2 j |, ∀j, k ∈ Kj , τ (Anderson et al., 1992; Seim, 2006). I check this
P
is l6=k∈Km | ∂akτ ∂alτ |
kτ

condition at the estimated equilibrium, finding that it is met for 99.9% of the observations.

The first of three sets of moment conditions equates the predicted mean utility for each station,

δˆkτ , found using the BLP contraction mapping, to its model prediction, which depends on the

computed advertising levels. These are

 
ω1kτ (θ) = Q0kτ δˆkτ − Xk β̄ + γ aˆkτ − ξkτ ∀k, τ (11)

where δˆkτ and aˆkτ are solutions to the contraction mappings, θ ≡ {β̄, Λ, σ, γ, ρ, α} is the vector of

all parameters, and Qkτ is a vector of instruments.

Because the amount of variation across markets in some demographics is quite small, I also

17
include an additional set of moment conditions, as in Petrin (2002). These match the predicted

total format demand conditional on a demographic characteristic to that reported by Arbitron

(2004). For example, the gender moment condition for f = Rock is

P P P RR
m k∈Kf τ Dikτ dPZ male dPv M enm
ω2 (θ) = P P P RR − sf,male (12)
m k∈Kf τ Dikτ dPZ dPv popm

where Kf is the set of all Rock stations, PZ male indicates the distribution of demographics for men,

M enm is the number of men in the market, and sf,male is the percentage of all Rock listeners who

are male. The same process is used for three age groups, black audiences, Hispanic audiences, and

college graduates with several formats.

The final set of moments matches the observed to the model predicted market advertising prices.

SRDS reports SQAD, Inc.’s weighted average market advertising prices for each day-part in a cost-

per-ratings point (“CPP”) measure. The CPP is the price per minute of advertising per one percent

of the market population. The SQAD weights are based on a proprietary algorithm intended to

create a consistent estimate of the average cost of advertising in the market. The average cost of

advertising in this model is calculated by using the quantity of advertising exposures to listeners

for each station in each day-part as weights. The moments are based on a transformation of the

CPP data into a price per listener. They are

P
k∈Km πkτ Dkτ akτ
 
CP Pτ m
ω3mτ (θ) = Q0mτ − P ∀m, τ, (13)
1% · popm k∈Km Dkτ akτ

where Qmτ is the market average of the instruments, Qkτ , above.

The Generalized Method of Moments objective function is

ω1 (θ)0 Ω̂−1 0 0 −1
1 ω1 (θ) + ω2 (θ) ω2 (θ) + ω3 (θ) Ω̂3 ω3 (θ), (14)

where Ω̂ is a consistent estimate of E[Q0 ωω 0 Q]. The instruments, Qkτ and Qmτ , include the

station characteristics, market population and demographics, two measures of market structure:

18
the percent of stations in the same format and the percent of stations owned by the firm, the

number of retail establishments in the market, and appropriate interactions. Instruments increase

the efficiency of the GMM estimator and allow for consistent estimation with endogenous market

share and price data in the first-order conditions. See Appendix C for more details.

7 Results
7.1 Parameter Estimates

The subset of estimated parameters describing the role of station characteristics in listener demand,

are presented in Table 4. The estimates of β̄ show that stations with more powerful signals, more

established identities, and with FM service have larger audiences. The artists of the second most

common format, Adult Contemporary, are the most popular. “AC” artists are also the most

commonly heard on other formats. On average, listeners of most demographic groups prefer Adult

Contemporary and Urban artists to other programming. Contemporary Hits and Other are the

least popular on average. The negative parameter estimates indicate that the listeners’ outside

option, not listening to commercial radio, is preferred to all formats. This is not surprising since

the share of the outside option is at least eighty-five percent for all markets. The standard deviation

of the preference for commercial radio, σ, is small relative to β̄. A Lagrange Multiplier test could

not reject the null hypothesis that σ is the same for all programming formats.

On the other hand, the deviations from mean utility, µ, are large relative to β̄. The second

column of Table 4 presents the estimates of Λ, the parameters of the random listener demographic

component of utility. Preferences for station characteristics vary significantly with age, gender, and

ethnicity. Preferences for the outside good vary with education, commuting time, and region. Pref-

erence rankings change with demographics, implying that audience substitution patterns, among

stations and the outside good, differ from market to market with the distribution of demographics

and station characteristics.

19
 Table 4: Structural Model Estimates, Listener Preference Parameters

Structural Parameter Estimate Std. Err. Structural Parameter Estimate Std. Err.

β̄ Antenna Power (kW) 1.00 4.E-03 Λ Age∗Rock -0.01 1.E-03

Station 2+ Years Olda 0.21 9.E-04 Age∗Adult Contemp. 0.02 0.01
FM Frequency Band 0.71 1.E-03 Age∗Contemp Hits -0.28 0.03
Talk Format -7.60 0.09 Age∗Other (Classical) 0.35 5.E-03
Country Format -6.96 0.07 Age∗Oldies 0.04 1.E-03
Rock Format -9.22 0.73 Male∗Talk 3.09 0.11
Spanish Lang. Format -7.96 0.07 Male∗Rock 6.44 1.43
Adult Contemp. Format -3.11 0.17 Male∗Adult Contemp. -2.30 0.19
Religious Format -7.49 0.06 Black∗Country -2.05 0.02
Contemp. Hits Format -11.06 0.65 Black∗Religious 1.19 0.01

20
Other (Classical) Format -35.21 0.35 Black∗Urban 1.81 0.02
Urban Format -4.60 0.11 South∗Country 0.52 2.E-03
Oldies Format -7.00 0.08 Hispanic∗Spanish Lang. 1.75 0.01
College∗Comm. Radio -0.12 0.01
σ Standard Deviation 0.15 0.06 Commute Time∗Comm. Radio 0.95 0.04
of Radio Preference Evening∗Comm. Radio -1.29 5.E-03
Northeast∗Comm. Radio -1.59 0.01
Midwest∗Comm. Radio -1.02 0.01
West∗Comm. Radio -1.49 4.E-03
Northeast∗Power 1.70 0.01
Midwest∗Power 1.32 0.01
West∗Power 1.57 5.E-03
NOTE: Format characteristics are defined by the extent of artist overlap among them. There are 21,682 observations.
a Dummy variable for the same format and call sign for two years.
 Table 5: Structural Model Estimates, Advertising Parameters
Std.
Structural Parameter Estimate Error
ρ Ad Price Constant 3.71 0.26
Log Age -0.72 0.06
Male -0.07 0.07
Black 0.58 0.04
Hispanic -0.01 0.02
College 0.29 0.04
Earnings 0.66 0.06
Evening -0.32 0.02

α0 Market Ad Exposuresa 0.07 0.01

α1 Station Ad Congestionb 0.17 0.02

γ Ad disutility 0.15 0.02

NOTE: Price is measured in cents per ad exposure.

The parameters associated with advertising and prices are presented in Table 5. The 0.15

estimate of γ demonstrates listeners’ net-distaste for commercial breaks. At an average of three

minutes of advertising per quarter hour, the marginal effect of one more minute of advertising is

similar in magnitude to the difference in listener utility between FM and AM stations. The estimate

of α1 implies that each additional minute of ads aired per quarter hour lowers price per listener by

0.17 cents, and, according to the estimate of α0 , the price per listener declines by 0.07 cents for each

million listener-minutes of advertising exposure in the market. The implied average price elasticity

of market ad exposures is approximately -1.2, while the average price elasticity of a station’s own

advertising level is about -1.8.

The results confirm the high value of young, educated, higher income audiences to advertisers.

Advertisers pay a premium for black audiences, all else equal. They value black audiences with

their relatively homogeneous demographic profile, as show by Chandra (2009). This demographic

group’s strong radio listenership may also contribute to this result (Jordan and Beswick, 2001).

Prices fall during evening hours when advertiser (and audience) demand shifts strongly to television.

21
 Table 6: Predicted Advertising Minutes for the Average Quarter Hour
1998 2003
Day-part Mean Std. Dev. Mean Std. Dev.
Morning 3.09 0.31 3.08 0.28
Day 3.09 0.31 3.09 0.28
Afternoon 3.09 0.31 3.09 0.27
Evening 2.65 0.36 2.65 0.32

7.2 Predicted Advertising Levels

The predicted advertising levels fall into a fairly tight distribution from two to four minutes per

quarter hour. This tight distribution makes sense because competition for advertisers tempers

prices and thus the lower end of the quantity distribution. Competition for listeners who dislike

ads limits the high end of the distribution. Table 6 describes the predicted advertising time by

day-part. Note that advertising falls by about half a minute per quarter hour during the evening

when demand is low. The standard deviation is lower than would be observed for quarterly hour

observations because the model does not account for strategic timing of commercial breaks. It

assumes that stations spread advertising evenly over all quarter hours.

The distributions for 1998 and 2003 are similar. Thus, one cannot infer the relationship between

local market concentration and advertising by looking at the aggregate change over time. Much

of the variation in predicted station advertising levels is, instead, observed in the cross-section. A

simple regression of ad levels on market concentration reveals their correlation. The relationship is

given by

ln akτ = 0.97 − 0.03 ln stations + 0.05 ln HHI,

where HHI is the firm concentration index and stations is the number of stations in the market.

The R-square statistic for this regression is 0.24, and the HHI effect is significant at the 99% level

of confidence. An increase from the mean HHI of 56.6 by one standard deviation, 40.7, would

induce about a four percent increase in time spent on advertising broadcasts.

22
 Despite the aggregate positive correlation between advertising and market concentration, the

model predicts falling advertising levels with firm concentration in some markets. Figure 1 plots

the station average, standard deviation, minimum, and maximum estimated listener and marginal

revenue effects against local firm size. The listener effect of joint ownership consistently rises with a

firm’s market power, indicating notable listener substitution among stations. The marginal revenue

effect is usually smaller, but may be quite large in some cases. For thirteen percent of stations, the

marginal revenue effect outweighs the listener effect, and so predicted ad times are lower than if

the station were independent.

7.3 Counter-factual Demand Experiment

Market specific effects of market power over time are confirmed by a counter-factual experiment.

A counter-factual is necessary to control for changes in demographics and the outside option from

1998 to 2003, including growth in the Hispanic population, aging of the population, and competition

from new technology. This isolates the change in market structure, both consolidation under fewer

owners and programming format changes.

I begin by matching 1998 demographics and population levels to each metropolitan area’s corre-

sponding 2003 market structure for the 236 (of 263) markets observed in both years. An increased

value of the outside option, not listening to commercial radio, is reflected in lower station fixed

effects for 2003. To control for this, I rescale ξkτ for the 2003 stations to match the mean and stan-

dard deviation of the 1998 distribution. In this counter-factual, firms condition on their observed

2003 station formats and ownership and the 1998 market demographics and listener preferences. I

recompute equilibrium advertising levels and calculate station outcomes.

23
 Figure 1: Shifts in First-Order Conditions with Joint Ownership

% Shift due to Listener Effect

25%
Shift in Marg. Rev. Tuned Out with Joint Ownership

20%

15%

10%

5%

0%
1 2 3 4 5 6 7 8 9 10 11
Stations Owned in Market

% Shift due to Marginal Revenue Effect

25%
Max = 37%
Shift in Marginal Revenue with Joint Ownership

20%

15%

10%

5%

0%
1 2 3 4 5 6 7 8 9 10 11
Stations Owned in Market

NOTE: The dots indicate the means, the boxes span one standard deviation in each direction, and
the dashed lines span the full range of observations.

24
7.3.1 Advertising Analysis

In the counter-factual, average station revenues rise by 7.6% despite falling station ratings. Though

aggregate effects on ads are small once again, market-by-market results are more striking. In

Lafayette-West Lafayette, Indiana advertising rose by four percent, and predicted station revenues

rose by almost sixty percent. The increase in advertising amounts to about thirty more seconds of

advertising each hour on the average station. Lafayette exhibits the typical response to consolida-

tion in the data. The number of owners of twelve stations fell from ten to seven, inducing a change

in the concentration index near the median in the sample. The listener effect is over seventy times

larger than the marginal revenue effect for the average station in this market. Table 7 shows other

markets with larger increases in concentration and similar results, such as Peoria, IL, Mobile, AL,

and Terre Haute, IN.

On the other hand, advertising fell by six percent and revenues rose about thirty percent in

Minneapolis-St. Paul, Minnesota, where the marginal revenue effect and listening effects were of

similar magnitudes in 1998, and the marginal revenue effect, rose by twice as much. The outcomes

were similar in almost all the other markets with large marginal revenue effects. Most of these

markets were demographically diverse, with a percent black or Hispanic or aged over sixty-five

years well above the median. Because preferences are strongly driven by demographics, diversity

segments the audience, giving firms relatively greater market power over advertisers.

The model does not predict egregious exercises of market power from 1998 to 2003. This is not

surprising given that antitrust authorities screened all mergers and station acquisitions during this

time period and required hundreds of station divestitures. Table 7 lists some markets where the

Department of Justice required station divestitures with mergers. These show smaller changes in

the concentration index and advertising.

25
 Table 7: Market-by-Market Estimated Average Advertising and Revenue Changes for the Counter-factual Experiment
Listening/ %Black %Age
Market %∆ Ads %∆ Revenue %∆ HHI MR Effect or Hispanic 65+

Lafayette-W. Lafayette, IN 4.30 58.9 73.0 77.0 6.8 8.4

Markets with Large Concentration Increases
Fort Pierce, FL -0.79 41.8 190.7 7.9 17.1 22.0
Springfield, MO -0.47 31.2 150.6 21.9 0.0 38.6
Mobile, AL 0.25 46.2 137.2 14.7 21.5 12.1
Terre Haute, IN 2.26 67.6 160.6 50.9 5.1 12.1
Peoria, IL 3.61 32.5 161.5 21.8 5.4 20.4
Markets with Large MR Effects
Minneapolis-St. Paul, MN -3.12 29.3 75.9 0.6 4.9 9.5
Memphis, TN-AR-MS -2.89 15.3 76.8 3.3 48.7 13.7
St. Louis, MO-IL -2.33 11.9 49.3 0.6 16.6 17.2
Charlotte-Gastonia-Rock Hill, SC -2.27 10.2 17.8 2.1 27.5 16.5
Atlanta, GA -1.57 -14.1 45.5 0.3 31.3 10.5
Las Vegas, NV -1.45 28.1 109.2 2.3 22.8 11.0

26
Phoenix, AZ -1.40 15.7 65.6 0.4 26.0 13.6
Pittsburgh-Beaver Valley, PA -1.27 5.3 23.1 0.6 6.5 23.3
Columbus, OH -1.22 15.6 48.5 2.0 14.0 13.1
Norfolk-VA Beach-Newport News, VA -0.67 26.6 60.5 1.5 33.1 12.0
San Antonio, TX 1.02 22.3 -0.1 1.5 57.4 16.0
Kansas City, MO-KS 2.25 4.9 -26.6 1.4 11.9 16.4
Markets with Large MR Effects and DOJ Divestitures a
Tampa-St. Petersburg-Clearwater, FL -1.91 -11.8 -26.3 0.6 20.1 27.7
Orlando, FL -0.54 -3.9 -29.2 2.1 21.6 14.8
Other Markets with DOJ Divestitures a
Louisville, KY-IN -1.28 -2.1 -5.1 3.8 10.1 15.3
Fort Myers-Cape Coral, FL -0.61 10.5 -26.3 7.0 14.2 21.6
Wichita, KS 0.54 15.0 66.2 8.6 10.9 12.4
Youngstown-Warren, OH-PA 0.83 9.7 125.4 7.8 4.5 18.1
Dayton-Springfield, OH 0.88 -2.4 16.8 3.2 13.9 17.8
NOTE: Model predictions are quarterly hour estimates of advertising and revenue averaged across day-parts.
1998 Demographics - median %Black or Hispanic: 14, median %Age 65+: 15, median %∆ HHI: 22.
a Department of Justice (1999)
7.3.2 Welfare Analysis

Shifting demand for radio between 1998 and 2003 led to a decrease in listener welfare of about

two percent for all demographic groups and a subsequent decline in revenues for broadcasters.

Because listeners do not pay, consumer surplus cannot be measured in dollars. Panel A of Table

8 presents the changes in listener utility by demographic group. It fell more for men, non-blacks,

and Hispanics. Black listeners experienced the only increase in surplus from radio, due to the

large expansion of the Urban format. During this time period, the Hispanic proportion of the

population in the average market rose by two percentage points. Hispanic audience shares rose less

than proportionately to the population increase. Despite large increases in the number of Spanish

language stations, sixty-five percent of markets still had no commercial Spanish language station

in 2003. Thus, the average Hispanic listener experienced the sharpest decline in utility from radio.

There is little difference between the welfare changes by age and education compared to the average

listener.

A combination of two factors affect the outcomes described above: 1) changes in the nature

of competition due to consolidation and 2) a shift in underlying market demographics. Panel B

of Table 8 presents the welfare results for the counter-factual experiment described above, which

controls for changes in demographics and the outside option. If the loss of commercial radio listeners

to the outside good had not occurred, station ratings would have remained fairly stable. The

counter-factual analysis does not take into account any changes in market structure between 1998

and 2003 that may have occurred in response to demographic shifts. For the sake of comparison,

I focus on the results for a subset of twenty-seven markets in which the least demographic change

or population growth occurred. They include Pittsburgh, PA, Wichita, KS, Lansing-East Lansing,

MI, Redding, CA, Buffalo-Niagara Falls, NY, and Jackson, MS among others. Panel C of Table 8

presents the welfare outcomes for this subset of markets.

27
 Table 8: Predicted Listener and Broadcaster Welfare
(A) (B) (C)
Base Counter-factual Counter-factual
All Observations All Observations Small Demographic Changes
1998 2003 %∆ 1998 CF %∆ 1998 CF %∆

Listener Welfare by Demographic

· All Audiences 0.740 0.726 -1.93 % 0.740 0.754 1.88 % 0.745 0.769 3.32 %
· Male 0.784 0.764 -2.65 % 0.784 0.799 1.83 % 0.796 0.822 3.32 %
· Female 0.701 0.690 -1.44 % 0.701 0.714 1.92 % 0.701 0.725 3.31 %
· Black 0.742 0.748 0.77 % 0.742 0.782 5.44 % 0.735 0.797 8.48 %

28
· Non-Black 0.740 0.723 -2.28 % 0.740 0.750 1.33 % 0.746 0.764 2.40 %
· Hispanic 0.766 0.736 -3.89 % 0.766 0.788 2.87 % 0.783 0.800 2.15 %
· Non-Hispanic 0.739 0.725 -1.81 % 0.739 0.752 1.80 % 0.743 0.768 3.38 %
· College Graduate 0.726 0.711 -2.05 % 0.726 0.733 0.94 % 0.728 0.742 1.96 %
· Non-College Graduate 0.745 0.731 -1.88 % 0.745 0.761 2.16 % 0.749 0.776 3.65 %
· Under 40 years old 0.751 0.736 -1.94 % 0.751 0.760 1.19 % 0.754 0.774 2.67 %
· Over 50 years old 0.728 0.714 -1.89 % 0.728 0.749 2.82 % 0.735 0.766 4.28 %
Broadcasters
· Station Ratings 0.70 % 0.59 % -15.34 % 0.70 % 0.68 % -2.09 % 0.70 % 0.71 % 1.94 %
· Revenues per hour $ 433 $ 404 -6.65 % $ 433 $ 465 7.60 % $ 471 $ 505 7.27 %
NOTE: Listener welfare measures are averages for the markets observed in both years.
 The popularity of format changes in 2003 outweighed the effects of advertising for the average

listener. The larger effect in Panel C suggests that these format changes are associated with the

higher ownership caps as opposed to demographic trends. The results in Panels B and C imply

that black listener welfare increases most of all demographic groups with consolidation. They also

show that Hispanic listeners benefit, especially in markets with more demographic change. Perhaps

higher ownership caps allow Spanish language stations to share some costs within a larger firm.

Younger listeners and men enjoy commercial radio more than the average listener, but they do not

experience greater increases in welfare due to consolidation. In fact, older listeners benefit a bit

more than younger listeners.

8 Conclusion

This paper presents a structural empirical model of the radio industry that can be used to quantify

strategic incentives and welfare implications of mergers in a two-sided market. The results show

that while commercial radio stations lost listeners and revenue from 1998 to 2003, increased radio

ownership caps allowed them to strategically adjust advertising levels to increase revenues relative

to what they would have been without consolidation. Accompanying changes in programming

formats provided modest net-benefits to audiences.

Antitrust authorities blocked potential mergers that would have had large adverse effects on

advertisers or audiences. However, some markets appear to have seen notable increases or decreases

in advertising. The market-by-market counter-factual analysis reveals that in most markets audi-

ences are at risk of increased advertising with local radio consolidation, while in more segmented

markets, advertisers would face higher prices. As demographic diversity continues to grow across

the U.S., the latter case may become more common. The insights and methods provided here may

be useful to policy makers evaluating media mergers and determining ownership policy.

29
 Much of the research on media ownership caps, both in economics and other disciplines, focuses

on the impacts on public access to the media, local content, and other public interest issues.

This paper is not able to address these issues, but it shows important discrepancies in welfare

outcomes across markets. Any adverse outcomes of media consolidation not addressed here may

also disproportionately affect certain audiences or cities despite small changes in the aggregate.

References

Anderson, S. P. and Coate, S. (2005), “Market Provision of Broadcasting: A Welfare Analysis,”

The Review of Economic Studies, 72, 947–972.

Anderson, S. P., de Palma, A., and Thisse, J.-F. (1992), Discrete Choice Theory of Product Differ-

entiation, MIT Press.

Arbitron (2004), “Radio Today: How America Listens to Radio,” http://www.arbitron.com.

Argentesi, E. and Filistrucchi, L. (2007), “Estimating market power in a two-sided market: The

case of newspapers,” Journal of Applied Econometrics, 22, 1247–1266.

Berry, S. T. (1994), “Estimating Discrete-Choice Models of Product Differentiation,” RAND Jour-

nal of Economics, 25, 242–262.

Berry, S. T., Levinsohn, J., and Pakes, A. (1995), “Automobile Prices in Market Equilibrium,”

Econometrica, 63, 841–890.

Berry, S. T. and Waldfogel, J. (1999), “Free Entry and Social Inefficiency in Radio Broadcasting,”

RAND Journal of Economics, 30, 397–420.

— (2001), “Do Mergers Increase Product Variety? Evidence from Radio Broadcasting,” The Quar-

terly Journal of Economics, 116, 1009–1025.

30
Brown, K. and Williams, G. (2002), “Consolidation and Advertising Prices in Local Radio Markets,”

Media Ownership Working Group Federal Communications Commission.

Chandra, A. (2009), “Targeted Advertising: The Role of Subscriber Characteristics in Media Mar-

kets,” Journal of Industrial Economics, 57, 58–84.

Chandra, A. and Collard-Wexler, A. (2009), “Mergers in Two-Sided Markets: An Application to the

Canadian Newspaper Industry,” Journal of Economics and Management Strategy, 18, 1045–1070.

Chipty, T. (2007), “Station Ownership and Programming in Radio,” Research Studies on Media

Ownership, Federal Communications Commission.

Department of Justice (1999), “Justice Department Approves Clear Channel’s Acquisition of Jacor

Communications After Parties Agree to Divestitures,” Press Release.

Fan, Y. (2010), “Ownership Consolidation and Product Quality: A Study of the U.S. Daily News-

paper Market,” Working Paper, University of Michigan.

Jordan, R. and Beswick, P. (2001), “African-Americans Growing in Affluence; Remain Most Loyal

Radio Audience in U.S.” The Media Audit.

Petrin, A. (2002), “Quantifying the Benefit of New Products: The Case of the Minivan,” Journal

of Political Economy, 110, 705–729.

Petrin, A. and Train, K. (2010), “A Control Function Approach to Endogeneity in Consumer Choice

Models,” Journal of Marketing Research, 47, 3–13.

Romeo, C. J. and Dick, A. (2003), “The Effect of Format Changes and Ownership Consolidation

on Radio Station Outcomes,” Review of Industrial Organization, 27, 351–386.

31
Ruggles, S., Sobek, M., Alexander, T., Fitch, C. A., Goeken, R., Hall, P. K., King, M., and

Ronnander, C. (2004), Integrated Public Use Microdata Series: Version 3.0 [Machine-readable

database], Minnesota Population Center [producer and distributor], Minneapolis, MN.

Rysman, M. (2004), “Competition Between Networks: A Study of the Market for Yellow Pages,”

Review of Economic Studies, 71, 483–512.

— (2009), “The Economics of Two-Sided Markets,” Journal of Economic Perspectives, 23, 125–144.

Seim, K. (2006), “An Empirical Model of Firm Entry With Endogenous Product-Type Choices,”

The RAND Journal of Economics, 37, 619–640.

Stuhmeier, T. and Wenzel, T. (2010), “Regulating Advertising in the Presence of Public Service

Broadcasting,” Working Paper, Dusseldorf Institute for Competition Economics.

Sweeting, A. (2007), “Dynamic Product Repositioning in Differentiated Product Industries: The

Case of Format Switching in the Commercial Radio Industry,” Working Paper w13522, National

Bureau of Economic Research.

— (2010), “The Effects of Mergers on Product Positioning: Evidence from the Music Radio Indus-

try,” RAND Journal of Economics, forthcoming.

Thomadsen, R. (2005), “The Effect of Ownership Structure on Prices in Geographically Differen-

tiated Industries,” RAND Journal of Economics, 908–929.

Wilbur, K. (2008), “A Two-Sided, Empirical Model of Television Advertising and Viewing Mar-

kets,” Marketing Science, 27, 356–378.

32
 Table A.1: Station Ownership Caps
Pre-1992 1992-1996 1996-Present
Stations Total AM/FM Total AM/FM Total AM/FM
Less than 15 2 1 3 2 5 3
15-29 2 1 3 2 6 4
30-45 2 1 4 2 7 4
45+ 2 1 4 2 8 5
NOTE: Information from FCC MM Docket No. 01-317

A Local Radio Ownership Rule

The Local Radio Ownership Rule was first established in 1941. It has been modified over time,

most recently with the 1996 Telecommunications Act, raising the limit on the size of a station

group from one AM and one FM station per owner to up to eight stations, with at most five in AM

or FM. When enforcing the Local Radio Ownership rule between 1997 and 2003, the FCC defined

markets on a case-by-case basis, depending on the signal contours of the stations involved in the

merger or acquisition. Since 2003 the rule has been applied using Arbitron’s market definitions.

All media ownership rules are reconsidered bi-annually as directed by the 1996 Telecommunications

Act. As of June 2010, details are provided at http://www.fcc.gov/ownership/.

B Using Music Artists to Define the Overlap Among Program-

ming Formats

Table B shows a small subset of the artists in the Mediabase 24/7 data for the year 2001. I

calculate each artist’s percentage of the song plays by the top 500 artists in each format. Note that

the denominator here is the total number of music plays for all stations in the format, not the total

number of all plays for the artist. For example, Alicia Keys songs accounted for 2.3% of the songs

played in the Urban format, which is higher than her percentage for any other format. Therefore,

she is defined as an Urban artist.

Using the artist assignments, I create a matrix of format overlap where the rows are radio

33
stations grouped by format and the columns are the artists’ assignments. Therefore, the cell

Country-Rock in Table B shows that 0.18% of songs played on Country stations are by artists

assigned to the Rock format. On the other hand, the Rock-Country cell indicates that only 0.13%

of songs played on Rock stations are by Country artists. The format overlap measures are product

characteristics in the listeners’ utility functions.

C Estimation Details

The mean utility, δkτ , from the BLP contraction mapping is not used as an input in the contraction

mapping over the first-order conditions. If it were, the changing ad values would not enter through

the market shares, which would imply a profit function that is not concave in akτ . This necessitates

an updating process for the station/day-part fixed effects, ξkτ , to minimize δˆkτ − Xkτ β̄ + γ aˆkτ .

The industry mean ad level is normalized through a penalty term in the objective function.

It does not affect the contraction mapping or the moment conditions. I use an average of three

minutes per quarter hour as cited by Sweeting (2010).

I minimize (14) over θ using the Nelder-Mead simplex search method. Following Nevo (2001),

the estimation begins using the product of the instruments, Q0kτ Qkτ and Q0mτ Qmτ to approximate

the weighting matrices Ω1 and Ω3 , respectively, and initial values of ξkτ = 0, ∀k, τ . With initial

convergence of the parameters, I update ξkτ for all stations and day-parts and repeat the parameter

estimation process. Finally, the Ω matrices are also updated, controlling for correlation of the errors

across day-parts and over time.

34
 Table B.1: Defining Format Overlap by Playlist Artists using 2001 Mediabase 24/7

Panel A: Artists’ Plays as a Percentage of All Plays by Format

Adult Contemp.
Artist Country Rock Spanish Contemp. Religious Hits Other Urban Oldies Assigned Format
Alicia Keys 0.006 0.020 0.163 0.965 0.246 2.335 Urban
Allman Brothers 0.001 0.210 0.035 Rock
David Allan Coe 0.032 Country
Enrique Iglesias 0.465 3.356 1.149 0.009 Spanish
Kenny G 1.144 0.016 2.287 0.008 0.030 Other
Third Day 0.031 0.016 3.470 Religious
U2 0.944 0.032 0.637 0.147 0.280 Rock

35
Panel B : Radio Format Playlist Overlap Characteristics
Percent of songs by artists most often played in category
Adult Contemp.
Station Category Talk Country Rock Spanish Contemp. Religious Hits Other Urban Oldies

Talk 100
Country 99.43 0.18 0.01 0.21 0.06 0.01 0.10
Rock 0.13 78.91 0.03 6.87 0.19 5.63 0.70 0.25 7.31
Spanish 3.60 92.31 0.36 1.14 1.35 1.23 0.00
Adult Contemp. 4.38 7.35 0.87 57.78 1.01 12.85 5.69 1.57 8.49
Religious 0.15 0.06 0.03 0.86 98.28 0.27 0.20 0.05 0.11
Contemp. Hits 0.11 6.29 0.22 7.98 0.35 66.61 0.07 18.31 0.06
Other 0.00 1.18 0.04 19.42 0.32 0.71
Urban 0.01 0.08 0.12 0.77 0.28 16.70 4.79 75.08 2.18
Oldies 0.36 5.19 0.07 11.94 0.11 0.04 2.87 1.28 78.13