Trash Media: How Competition Affects Information∗
Julia Cagé
Harvard University
November 18, 2012
Abstract
This paper questions the common wisdom whereby more competition in the media
industry leads necessarily to more information. I construct a voting model in which
individuals with heterogeneous tastes for “information” and “entertainment” vote strategically. I investigate the impact of a change in the intensity of competition in the media
market on the provided quantity of information and, through this channel, on electoral
turnout. I find that when newspaper buyers differ little in their taste for information and
the relative cost of producing information is high enough, more newspaper competition
leads to less information and a decrease in turnout. I confirm these predictions empirically
using a new panel of local daily newspapers and turnout at local elections in France from
1945 to 2011. I find that an increase in newspaper competition has a robust negative
impact on turnout at local elections. I next enter into the black-box of news-making to
explain this finding. I establish that, due to an important business stealing effect, an
increase in competition leads to a decrease in incumbent newspapers’ operating expenses,
and in particular the number of journalists. Through this channel, I find that an increase
in competition leads to (i) a lower provision of total news and, within these news, (ii) a
lower share of information and a higher share of entertainment. I finally show that more
competition leads to an increase in newspaper differentiation.
JEL No. D72, L82.
∗
I gratefully acknowledge the many helpful comments and suggestions from Alberto Alesina, Charles Angelucci, Daniel Cohen, Nathan Nunn, Dorothée Rouzet, Valeria Rueda, Andrei Shleifeir, Michael Sinkinson,
Nathalie Sonnac and James Snyder. This paper owes much to François Keslair who took an active part in
the construction of a first version of the dataset. I am also grateful to seminar participants at Harvard, the
Media Economics Workshop, MIT, NYU and PSE for input on this project. I would like to thanks all the
people and institutions that give me access to their data, especially the archivists of the French National
Archives. Cécile Alrivie, Danamona Andrianarimanana, Guillaume Claret, Georges Vivien Houngbonon, Romain Leblanc, Eleni Panagouli, Juliette Piketty, Graham Simpson, Kyle Solan and Alexander Souroufis provide
outstanding research assistance. This research was generously supported by the CEPREMAP, PSE and the
LEAP at Harvard. I also need to thanks the Center for European Studies for a Krupp Foundation Fellowship.
All errors remain my own. Email: cage@fas.harvard.edu.
1
”Half of the American people have never read a newspaper. Half never voted
for President. One hopes it is the same half.” (Gore Vidal)
1
Introduction
This paper questions the common wisdom whereby more competition in the media industry is necessarily leads to more information. More competition is often seen as implying an
increase in the dissemination of information, thereby enhancing the extent of ideological diversity. Indeed media competition, by raising the competitiveness of the marketplace of ideas,
is supposed to contribute to the political process. In this spirit, studies in political economy
have advanced that there exists a positive link between media competition and political participation. Be that as it may, there is however concern that the quality of the media might
have diminished during the last decades despite the manifest increase in the number of media
available. Indicative is the contemporaneous increase in the number of television channels
and the often perceived decrease in quality, exemplified by the multiplication of tabloid talk
shows and emergence of ”Reality” television.
In this paper I combine a voting model with a multidimensional differentiation competition
model to investigate the impact of a change in the intensity of competition in the media market
on the quantity of information provided and, through this channel, on political participation.
Information is defined as “accountability” or “fact-based” news, as opposed to entertainment,
defined as “commodity news” or “tabloid journalism”. On the supply side, profit-maximizing
newspapers choose the quantity of information and the quantity of entertainment they want
to produce, as well as their price. I first assume that there is only one newspaper in the
market (monopoly) and next study the impact of the introduction of a second newspaper
(competition). On the demand side, newspaper buyers differ in their tastes for information
and entertainment. They are also potential voters who behave strategically. The effect of
newspapers on voting is thus a by-product: voters information depends on the structure of
the news market and this information affects their participation. I find that when (i) the cost of
producing information is higher than the cost of producing entertainment and (ii) newspaper
buyers differ little in their taste for information, then newspaper competition leads to lower
information and, through this channel, lower political participation than monopoly.
I confirm these predictions empirically using a new panel of local daily newspapers and
turnout at local elections (city level) in France from 1945 to 2011. The choice of studying the
French news market is driven by the quality of the data I was able to collect.1 My dataset
1
The media literature relies mainly on American newspapers, with some exceptions: e.g. Di Tella and
Franceschelli (2009) on Argentina; Durante and Knight (2009) on Italy; Della Vigna, Enikolopov, Mironova,
Petrova, and Zhuravskaya (2010) on Serbia and Croatia; Enikolopov, Petrova, and Zhuravskaya (2011) on
Russia. To the extent of my knowledge, this paper is the first to construct a panel dataset on French media.
2
includes every local daily newspaper published in France over this time period. I observe
papers’ location, circulation, readership as well as, importantly, number of employees, cost
and revenue.2 I supplement this data with different measures of newspaper content described
in more details below.
Following Gentzkow, Shapiro, and Sinkinson (2011b), my basic strategy is to look at
changes in political participation in cities that experience a newspaper entry or exit relative
to other cities in the same region and year that do not. I find that newspaper competition
has a robust negative effect on turnout. One additional newspaper decreases local turnout
by approximately minus 0.4 percentage points. When considering only newspaper entries, I
find that the effect of an entrant on the market is minus 1.1 percentage points. This effect is
robust to a range of alternative specifications and controls. Moreover using the timing of the
change in the newspaper markets, I undertake a falsification test which assesses the validity of
my results: I find no impact of a future change in newspaper competition on current turnout.
This suggests that changes in the number of newspapers are not driven by election results
and brings confidence in interpreting my effects as being causal.
I next enter into the black-box of news-making to explain this finding. I first establish
that there is an important “business-stealing effect”: the entry of a newspaper reduces the
circulation of incumbent newspapers by 25%. Due to this business-stealing effect, the entry of
a newspaper leads to a decrease in both the revenues and operating expenses of incumbents.
In particular, the entry of a newspaper leads to a decrease in the number of journalists working
for incumbents. Moreover, there is no overall increase at the news market level: the aggregate
number of journalists working in a county (a news market) does not increase with the number
of newspapers in this county.3 I finally show that, through this “journalist” channel, more
competition leads to (i) a lower provision of total news and, within these news, (ii) a lower
share of information and a higher share of entertainment. That is, if the same number of
journalists is divided into two newspapers, and each try to cover all the news, then they both
tend to produce lower quality news. This is a simple but powerfull mechanism that help
understanding how competition can hurt news-making.
The quantity of total news produced is measured using different indicators: (i) the number
of words by front page articles, (ii) the total number of articles by newspaper issue and (iii)
the total number of words by newspaper issue. I find that a one-standard deviation increase
in the total number of employees working for a newspaper increases the number of articles
2
To give a flavour of what is generally available in terms of newspaper cost and revenue data, it is worth
remembering that in their study of how economic incentives shape ideological diversity in the media, Gentzkow,
Shapiro, and Sinkinson (2011a) have no other choice but to use balance sheet data on 94 anonymous newspapers
that they match with newspapers using circulation value. On the contrary, I have annual balance sheet data
for 49 newspapers from 1984 to 2009.
3
By a small abuse of language and for the sake of clarity, I use in this article the term “counties” when
refering to French “departments”. Similarly, I use the term “states” when refering to “regions”. French local
jurisdiction are described in more details in the Appendix (Appendix Table 9).
3
per newspaper issue by nearly 0.7 standard deviation.
Within these news, I then study the distribution of articles across topics, separating articles on information (e.g. on economics or politics) and articles on entertainment (e.g. on
sports or leasure activities). I find that an increase in competition leads to a decrease in
the share of articles on information (and, symmetrically, an increase in the share of articles
on entertainment). I also find that a one-standard deviation increase in the total number
of employees working for a newspaper increases the share of articles on information by 0.3
standard deviation.
I finally show that more competition leads to more newspaper differentiation: an increase
by one in the number of newspapers in the news market leads to a 0.4 standard deviation
increase in the Herfindahl index of newspaper specialization.
Taken together, these results are consistent with the predictions of my model in which
(i) newspapers affect the political process primarily by providing information about elections
and (ii) this provision can be affected by change in the degree of competitiveness of the news
market.
Related Literature
There is a large political economy literature on how media affects
political participation, but its focus is much more on media access than on media competition.
Stromberg (2004b) finds that the introduction of the radio in the 1920s and the following rapid
increase in the share of households with radios led to more people voting in gubernatorial
races. Similarly, Oberholzer-Gee and Waldfogel (2009) find that the introduction of Spanishlanguage local television increases the turnout among Hispanics. Gentzkow, Shapiro, and
Sinkinson (2011b), using a panel of local US daily newspapers, show that one additional
newspaper increases turnout at national election but underline that the effect is ”driven mainly
by the first newspaper in a market”.4 My paper studies both theoretically and empirically the
non-monotonicity of this finding. Media access (going from 0 to 1 newspaper) can have very
different effects that changes in media competition (going from 1 to 2 newspapers, or from 2
to 3 newspapers). Given the dataset I am using, I am unable to capture the effect of the 0 to
1 margin (I always have at least one newspaper in a given news market). I am thus not able
to measure the effect of the first entrant but only the effect of later entrants. I find that this
effect is negative. This is consistent with the idea that an increase in the competitiveness of
the market may lead to a “race to the bottom” (Arnold, 2002).
The impact of competition on the revelation of information has also been the object of
a growing theoretical literature. Gentzkow and Kamenica (2012), assuming that information
is costless, find that competition among persuaders increases the extent of the information
4
See also Gentzkow (2006),Schulhofer-Wohl and Garrido (2009), Banerjee, Kumar, Pande, and Su (2010)
and Snyder and Stromberg (2010).
4
revealed.5 My framework is totally different from their – and much simpler – but it is worth
underlying that, once I assume that information is costly to produce, I obtain that more
competition decreases the provision of information.
A large theoretical literature also studies how media competition affects media ideological
bias. Mullainathan and Shleifer (2005) find for example that competition does not reduce and
might even exaggerate this bias (see also Besley and Prat (2006), Gentzkow and Shapiro (2006)
and Gentzkow, Shapiro, and Sinkinson (2011a)). But as underlined by Prat and Stromberg
(2011) in their recent survey of the literature on the influence of mass media, this literature
“leaves out an important body of research in industrial organization (...) that deals with the
media industry, mostly without any direct reference to the political system”. In this paper
I try to fill this gap by combining two classical building blocks: a political economy voting
system to study political participation and a multidimensional differentiation competition
model to determine newspapers provision of information. From this point of view, my paper
is closely related to Stromberg (2004a) who combines a model of mass media competition
with a model of political competition. However the model of media competition I develop is
quite different from his model. While Stromberg (2004a) constructs a horizontal competition
model – his focus is on the location of each newspaper in the news space –, my model is a
vertical competition model with multidimensional newspapers which vary in their price, the
amount of information and the amount of entertainment they provide. This allows me to
obtain new predictions on how newspapers competition may affect information.
From this point of view my paper contributes to the industrial organization literature
on how concentration affects differentiation and variety and in particular the literature on
mergers. Using data on reporter assignments from 1993 to 1999 George (2007) shows that
differentiation and variety increase with concentration in markets for daily newspapers. Berry
and Waldfogel (2001), using evidence from radio broadcasting, find similarly that increased
concentration increases variety absolutely.6 Focusing on the impact of market size on product
quality, Berry and Waldfogel (2010) show that in daily newspapers, the average quality of
products increases with market size, but the market does not offer much additional variety as
it grows large. They measure quality with the size (number of pages) of the paper and the
number of reporters on staff which are also my variables of interest. Interestingly their finding
is driven by the fact that the cost of quality is fixed with respect to output, which is an important assumption of my model. Even more closely related to my paper, Fan (2011) develops a
structural model of newspaper markets to analyze the effects of ownership consolidation and
simulates the effect of a merger in the Minneapolis newspaper market.
5
On the contrary, Angelucci (2012) shows that an increased misalignment of preferences amongst persuaders
may lead to less information revelation and more imprecise decisions if information acquisition is costly.
6
Other examples in this literature include Della Vigna and Kennedy (2011) who study media concentration
and bias coverage using movie reviews, and George and Oberholzer-Gee (2011) who examine the impact of the
local market structure on viewpoint diversity in the market for local television news.
5
My study differs from this past work in the large number of media outlets and the long
period of time it covers and from the links it establishes between the impact of competition
on the quality of information and political participation. To the extent of my knowledge, it
is the first large-scale empirical study to relate changes over time in news market structure
to changes in the cost and revenue structure of newspapers and to link these changes with
changes in political participation.
Finally, it is worth emphasizing that there is a growing empirical literature studying newspaper content but that the focus of this literature is on political bias and not on the quality
or quantity of information. Various measures of media bias have been used, in particular
measures of newspapers’ political leanings (endorsement, candidate mentions,...) using automated searches of news text. Groseclose and Milyo (2005) proxy the political positions of US
media outlets by the average ideology of the think tanks they quote. Gentzkow and Shapiro
(2010), exploiting the Congressional Record, use similarities between language used by media
outlets and congressmen. In this paper, I try to draw a new distinction between the share of
articles on information and the share of articles on entertainment newspapers produce.
Let me underline that, contrary to the existing literature, I assume in this paper that
there is no bias. I voluntarily choose to abstract from political bias considerations for two
reasons. First and most importantly, there is no political bias in French local daily newspaper
during my period of interest. As noted by Éveno and Des travaux historiques et scientifiques
(2003), since 1947, “the story of biased newspapers has been the one of a slow decline”. The
last local daily biased newspapers disappeared in France in the 1950’s. Second, it allows me
to keep the model tractable while identifying a new effect of newspaper competition on the
provision of information.
The remainder of the paper is organized as follows. Section 2 below lays out the model
and investigates the impact of competition on the provision of information and on political
participation. Section 3 describes the novel panel dataset of French local daily newspapers and
turnout at local elections used in the study. Section 4 lays out my empirical strategy, discusses
identification and presents my main results as to the impact of newspaper competition on
turnout at local elections. In section 5 I open the black-box of news-making to understand
why more newspaper competition leads to less information. Section 6 concludes.
2
The Model
My model combines two blocks together: a voting model and a differentiation competition
model. The voting part of the model is closely related to Feddersen and Pesendorfer (1996)
and Feddersen and Sandroni (2006a,b). Society much choose between two alternatives by
majority voting. There are two states of nature: one in which all voters prefer the first can6
didate and a second state where all prefer the other candidate. Voters have state dependent
preferences: there are no partisans. I voluntarily chose to abstract from political bias considerations. Readers do not have political opinions and individuals are only heterogenous in
their preferences for information and entertainment. Similarly, there is no media bias and
newspapers are pure profit-maximizers. Agents are motivated to vote out of a sense of ethical
obligation. Each agent has an action she should take and receives utility from taking this
action. Hence each agent behaves strategically even though pivotal probabilities play no role.
This is described in more details below.
2.1
2.1.1
Model Set-Up
Nature
There are two equally likely states of Nature Θ ∈ {0, 1} that are unobservable. There is
a continuum N of agents who share common prior about the state of Nature (one half).
There are two candidates running for the election, candidate 0 and candidate 1: Ω = {0, 1}.
The candidate that receives the majority of the votes cast is elected (if there is a tie, each
candidate is chosen with equal probability). One can think of the two candidates as being
the “status quo” and the “alternative”, and assume that there is some uncertainty about the
cost of implementing the alternative which can be either high or low.
2.1.2
Agents
Agents take two actions. First they choose whether to buy a newspaper and next they choose
whether to vote.
Newspapers Differentiation
Agents first choose whether to buy a newspaper: α ∈ A =
{B, N B} (B: buy; N B: do not buy). I assume that there is unit-demand: agents cannot
buy more than one unit of the newspaper. Moreover I assume that there is no multi-homing:
when there are two newspapers, agents can only buy one of the two. They cannot buy both
newspapers at the same time. This assumption is made to keep the model tractable and
allows me to rule out a potential channel through which newspaper competition could have
negatively affected the information received by the buyers, namely “confusion”. Indeed, under
certain conditions, too much information can decrease the quality of the information – the
signal – received by the readers.
Agent i maximizes the following utility function:

bi xj − pj ,
Vi =
0,
if she buys newspaper j
otherwise
7
(1)
where xj is the ratio of information over entertainment included in newspaper j and bi is
agent i’s taste for this ratio. I assume that this taste is uniformly distributed over the interval
b, b : U ∼ b, b . pj is the price of newspaper j. Agent i buys newspaper j iff bi xj − pj ≥ 0
and (if there is competition) bi xj − pj ≥ bi xk − pk , ∀ j 6= k.
Newspapers maximize their profits by choosing their price p and the ratio of information
over entertainment x they want to produce:
"
max
(xj ,pj )
cx2j
pj Dj (xj , pj ) −
2
#
(2)
where Dj (xj , pj ) is the demand for newspaper j given xj and pj , and c is the production
cost of x. Under duopoly, newspapers simultaneously choose x and then compete on price p.
Intuitively, one can think of x as the fraction of pages in the newspaper devoted to information
that is costly to produce.
I assume that the news market is one-sided, i.e. I do not take into account newspaper
dependency on ad revenues. I recognize that newspapers derive revenue from both readers and
advertisers. Implicitly here I am considering advertising revenue as a per-reader proportional
subsidy.7
Political Participation Agents next choose whether to vote: s ∈ S = {a, 0, 1}, where a
denotes abstention, 0 denotes vote for candidate 0 and 1 vote for candidate 1. As I underline
above, there is no partisans. Voters have state dependent preferences, i.e. given a pair (ω, θ),
ω ∈ Ω and θ ∈ Θ, the utility of a potential voter is:

0,
if ω 6= θ
U (ω, θ) =
U > 0, if ω = θ
Moreover, I assume that there is a uniformly distributed cost to vote C ∼ U 0, C .
Every voter receives a message m ∈ M = {0, 1, φ}. Voters who receive a message 0 or
1 are informed and all others are uninformed. I assume that the information acquisition is
exogenous in the voting stage of the game: voters who buy a newspaper are informed and all
others are uninformed. In other words in my setting, the effect of newspapers on voting is a
by-product.8
I call q ∈ (0, 1) the fraction of informed voters in the population. q =
D
N
where D is
the demand for the newspaper in the monopoly case, and the sum of the demands for both
7
In Angelucci, Cagé, and De Nijs (2012), we introduce the two-sidedness dimension of the news market to
investigate how newspapers change their price discrimination strategies between subscribers and other readers
as a function of their dependency to advertising revenues.
8
A possible extension will be to endogenize the acquisition of information. However it will make the model
much less tractable without modifying its main predictions.
8
newspapers in the duopoly case (remember that N stands for the size of the population).
Among the informed voters, the fraction which observes the message m ∈ {0, 1} in state m is
ρ ∈ (.5, 1]. When ρ is close to 0.5 the message is a very noisy signal of the true state, while
when ρ is close to 1 the message almost perfectly conveys the true state.
I assume that ρ is an increasing function of x s.t. ρ (0) = 0.5 and ρ0 (x) > 0. In other
words, the higher the ratio of information over entertainment provided by the newspaper, the
better the quality of the signal received by the reader.
2.1.3
Timing of the Game
The game proceeds as follows:
1. Nature draws θ ∈ Θ = {0, 1}.
2. Newspapers choose the ratio of information over entertainment x and the price p.
3. Voters choose α ∈ A = {B, N B} (whether to buy a newspaper, and which one).
4. Voters choose s ∈ {a, 0, 1} (voting decision).
5. The state of nature is revealed.
I solve the game by backward induction.
2.2
Solving the Model
Solutions fall under three cases, depending on the degree of differentiation of consumers taste
for the information over entertainment ratio. Solving the game by backward induction and
comparing what happens to the information over entertainment ratio under monopoly (x∗m )
with what happens under duopoly (x∗d,1 ,x∗d,2 ), I obtain the following proposition.
Proposition 1
If b > 4b (high taste differentiation), then x∗d,1 < x∗m < x∗d,2 .
If 2b < b ≤ 4b (intermediate taste differentiation), then x∗d,1 < x∗d,2 < x∗m .
If b ≤ 2b (low taste differentiation), then x∗d,1 = x∗d,2 =
x∗m
2
< x∗m .
Proof. See Appendix.
The impact of competition on information provision thus depends on the degree of taste
differentiation. Proposition 2 tells us how it impacts turnout.
9
Proposition 2 (Rational Abstention)
(i) Only informed voters (reading a newspaper) vote. Uninformed voters never vote.
(ii) Among informed voters, if there are different degrees of information – two newspapers
with different x competing on the market –, then only the informed voters reading the newspaper with the higher x vote.
(iii) There is cut-off point such that better informed voters with voting costs above this threshold should abstain. This cut-off point is increasing in x.
Proof. See Appendix.
Combining Propositions 1 and 2 I obtain the following predictions from the model.
Prediction 1 (High Differentiation) If there is high differentiation in voters taste for information, then:
(i) Turnout is higher under duopoly than under monopoly.
(ii) Voters are better informed under duopoly than under monopoly.
Prediction 2 (Intermediate and Low Differentiation) If there is intermediate or low
differentiation in voters taste for information, then:
(i) Turnout is lower under duopoly than under monopoly.
(ii) Voters are less informed under duopoly than under monopoly.
In the following sections, I test these predictions empirically.
Data Sources and Descriptive Statistics9
3
3.1
3.1.1
Newspapers Data
Number of Newspapers
To determine for each year between 1945 and 2011 the number of newspapers present in each
French county I use various sources of information that I digitize and merge together. This
dataset is the first dataset on the long-run evolution of the French news market. I choose the
county as my unit of analysis since it is the natural news market.
I count local daily newspapers from these sources: in each year, I extract the name and the
county(ies) in which circulates every local daily newspaper. I match newspapers across year
on the basis of their title and county(ies), allowing newspapers to circulate across counties.
For each county-year, I then compute the number of local daily newspapers which serves as
my key explanatory variable.
9
For the description and the sources of the data in more details see the Data Appendix.
10
The sample includes 288 newspapers. I observe a total of 266 county-years with net
newspaper entry and 340 county-years with net newspaper exit between 1945 and 2011. These
266 entries and 340 exits are key for my identification strategy.
Newspaper Owners A possible concern might come from the fact that the effect of an
increase in competition may be different whether the entrant newspaper is owned by the same
owner than the incumbent newspaper, or by a different owner. To determine the identity of
newspaper owners, I use several historical sources described in details in the Data Appendix.
My sample includes 278 owners. Over the period, I observe a total of 281 county-years with
net owner entry and 387 county-years with net owner exit.
3.1.2
Newspaper Circulation
For the period 1945-1990, newspaper circulation data comes mainly from archival data that
I digitize and merge together. Data for recent years (1990-2011) comes from the OJD (the
French press observatory whose aim is to certify the circulation data).
Between 1964 and 2011, I also have data on the geographical dispersion of circulation
when newspapers circulate across nearby counties. Having this data is important since even
if local daily newspapers in France are county-level newspapers, some of them circulate across
counties. Hence I do not want to bias the analysis by constraining the news market to be a
county and newspapers to circulate only in the news market where they are headquartered.
Doing so would have lead to an underestimation of newspaper competition.
3.1.3
Newspaper Readership
I collect annual data on newspaper readership by newspaper between 1957 and 2011 from
studies on French newspaper readers conducted principally in order to provide information to
advertisers. For the 1957-1992 period, I digitize data from surveys conducted by the CESP,
a French interprofessionnal association gathering the whole of the actors of the advertising
market concerned with the study of media audience. For more recent years I download all
the annual audience studies available in an electronic format at the French local daily press
syndicate which accepted to share this non-publicly available information with me. These
surveys mainly cover for each newspaper information on its aggregate readership. However,
for the sub-period 1996-2004 there is also information on readership by county for newspapers
circulating across nearby counties. Having information on both newspaper circulation and
readership allows me to compute the ratio of reported readership to circulation.
11
3.1.4
Newspaper Profitability and the Number of Journalists
Firm Data (1984-2009)
I compute annually for local daily newspapers between 1984 and
2009 a number of important economic indicators, namely sales, profits, value-added, operating
expenses (payroll, inputs, taxes), operating revenues (revenues from sales and revenues from
advertising), and the number of employees. The panel data covers 43 newspapers from 1984 to
2009, plus 5 other newspapers from 1993 to 2009. This data is from the (i) Enterprise Survey
of the French national institute for statistics which covers the period 1993-2009, and the files
constructed for the tax regime ”Bénéfice Réel Normal” (BRN) by the ”Direction Générale des
Impôts” (DGI) from 1984 to 2009. I identify newspapers in the dataset by using the French
registry of establishments and enterprises.This dataset is, to the extent of my knowledge, the
most complete existing dataset on newspapers cost and revenue.
Journalist Data One of the important downside of using French data is that, contrary
to what exists for example in the United States, there is no media directory available with
information on the number of journalists by newspapers. Hence I have to find other data
sources in order to be able to estimate how competition impacts the number of journalists.
I use two different sources: (i) information on the total number of employees working for
each newspaper that is provided in the firm survey described above; (ii) for more recent
years (1999-2011), I obtain data on the number of journalists (and on the total number of
employees) directly from the local daily press syndicate.This dataset is not complete (it has
information for only 24 newspapers – I complete it by data I obtain directly from newspapers
when available) but it allows me to compute the share of the journalists in the total number
of employees. This share is pretty stable during the 13 years for which I have data (37% on
average) (see Appendix Table 11). The correlation coefficient between the total number of
employees and the number of journalists is equal to 0.94 (and statistically significant at 1%).
3.1.5
Newspaper Content
I use various sources to study newspaper content. First, I use newspaper front pages and,
for each newspaper issue, count the number of words by front page. There are a number
of advantages of using front pages. First, front pages are available daily for 51 newspapers
over the period 2006-2012. I download them from the local daily press syndicate using an
automated script. Hence the panel data of front pages is very complete and it is balanced.
Second, as shown below, there is a strong correlation between the number of words on the
front page and the total number of words and articles inside the newspaper. Using front pages
is not new in the literature. To establish evidence of media capture, Di Tella and Franceschelli
(2009) construct an index of how much first-page coverage of the four major newspapers in
Argentina is devoted to corruption scandals.
12
Second, I obtain the entire daily content of each newspaper issue by using an automated
script to retrieve for each day and each newspaper issue all the articles published in the issue.
I obtain this information by downloading the information available on two different websites,
Factiva and Lexis-Nexis, which aggregate content from newspapers. I construct a dataset
covering 22 different newspapers over the period 2005-2012. I use this information to obtain
measures of the quantity of the total news produced: the total number of articles and the
total number of words per issue. Berry and Waldfogel (2010) also use the size of the paper
to measure quality, but they measure it with the number of pages, not with the number of
articles per issue.
I next use the metadata associated with each article on Lexis-Nexis (title, subject, topic)
to classify articles between “information” and “entertainment”. The share of articles on
“information” is defined as the number of articles on agriculture, economics, education, environnement, international or politics, divided by the total number of articles that I am able
to classify. The share of articles on“entertainment” is defined as the number of articles on
movies, culture, leisure activities, sports, “news in brief”, religion or health, divided by the
total number of articles that I am able to classify. (By construction, the sum of both shares
is equal to 100).
Finally, I use the article classification in sub-categories to construct a measure of newspaper differentiation. This measure is simply an Herfindhal index varying between 0 – no
specialization, i.e. no differentiation between newspapers that all deal with all the topics
– and 1 – perfect newspaper specialization, i.e. important newspaper differentiation, each
newspaper being specialized in a given topic (e.g. music or sport). This index is equal to
the sum of the squares of the shares of the different newpaper topics in each newspaper issue: agriculture, culture, economics, education, environnement, health, international, leisure
activities, movies, “news in brief”, politics religion and sports. I compute it both on a daily
(considering each newspaper issue separately) and or a weekly basing (summing all the issues
of each newspaper for each given week together).
3.2
Electoral Data
The main focus of this paper is on mayoral elections (city-level elections). Studying the impact
of local newspapers on participation at local (rather than national) election is an important
difference with Gentzkow, Shapiro, and Sinkinson (2011b) who study turnout at presidential
and congressional elections. However, since I am considering the impact of local newspapers,
I think that it is more appropriate to use local elections.
As of today, there are 36,570 communes in metropolitan France. I focus on the provinces
and do not take into account the “Paris area”.10 . There are 2,282 communes with more
10
In this area local daily newspapers are competing in a different way with national newspapers. Paris having
13
than 3,500 inhabitants in the provinces. In this paper I focus on these communes over 3,500
inhabitants since there is a change in the electoral rule for mayor elections for communes
under 3,500 inhabitants. For each election, I measure turnout as the ratio of cast votes to
eligible voters. I use cast votes rather than total votes since in France white votes are not
included in turnout.
Mayoral elections take place in France every six years. Between 1945 and 2010, 12 elections
took place (1945, 1947, 1953, 1959, 1965, 1971, 1977, 1983, 1989, 1995, 2001 and 2008). I
choose not to include in the dataset the turnout results for 1945 since this election took place
before the end of the Second World War in very special conditions and it happened just two
years before the 1947 election. Before 1983, data on French municipal elections have never
been digitized. Hence I construct the first electronically available dataset on French local
elections results at the commune level between 1945 and 1982, mainly from archival data.
More recent data are from the Centre de Données Socio-Politiques (CDSP) of Science-Po
Paris, the Interior Ministry, and Bach (2011).
3.3
Demographic Controls
City-level demographic data from the French census are in electronic format from 1962 to 2010.
I digitize data for the 1936, 1946 an 1954 censuses from books published by the INSEE. I
compute the share of the population that is 20 and older, the share of the population employed
in manufacturing and the share of the population having the baccalaureate or more. For each
measure, I interpolate both the numerator and denominator between census years using a
natural cubic spline (Herriot and Reinsch, 1973) and divide the two to obtain an estimate of
the relevant share.
4
Newspaper Competition and Electoral Turnout: Empirical
Evidence
4.1
Specification and Identification Strategy
I match my panel data on newspaper competition with mayoral election results from 1947 to
2008 and track the impact of a change in competition on turnout. Let C index cities, c index
counties and t ∈ {1, ..., 11} index election years (one time unit representing six calendar years).
The outcome of interest, yCt , is voter turnout. The key independent variable of interest is
NCt , which is defined as the number of newspapers in city C at time t.
I assume that
a national dimension, a lot of “local” information concerning the ”Paris area” is in fact taken into account in
national newspapers. Then there is much more competition between the different newspapers than in the rest
of France and considering only the competition between the local newspapers would be misleading.
14
yCt = αNCt + δxCt + ρC + µst + εCt
(3)
where ρC is a city fixed effect, µst is an election-state fixed effect, xCt is a vector of
observable characteristics, δ is a vector of parameters and εCt is a city-year shock. The
parameter α is the causal effect of NCt (the number of newspapers) on yCt . Since turnout
varies at the city level while the number of newspapers varies at the county level (if two cities
are in the same county, they have the same number of newspapers) I cluster the standard
errors at the county level.11
My identification relies on changes in the number of newspapers over time. As a result
it is correct as long as the timing of these change is random. I undertake a falsification test
using the timing of the change which seems to confirm that it is indeed the case. Similarly
to what is done in Gentzkow, Shapiro, and Sinkinson (2011b), I estimate the model in first
differences. I let ∆ be a first difference operator so that, for example, ∆NCt = NCt − NC(t−1) .
Unless otherwise noted, the vector of control xCt includes the share of the population that
is 20 and older, the share of the population employed in manufacturing and the share of the
population having the baccalaureate or more.
4.2
Main Results on Turnout
Table 1 presents my baseline results (all specifications include election-state fixed effects and
city-level demographic controls). Column 1 shows the effect of one additional newspaper on
local turnout: I find that one additional newspaper decreases turnout by approximately 0.3
percentage point. In column 2 I focus on newspaper entries. When considering only entries I
find that the effect of an entrant on the market is minus 1.1 percentage points.
A potential concern might come from the fact that the effect of an increase in competition may be different whether the entrant newspaper is owned by the same owner than the
incumbent newspaper or by a different owner. In Column 3, I focus on owner entries. When
considering only owner entries I find that the effect of an entrant on the market is minus 1
percentage points, i.e. of the same order of magnitude than the effect in column 2.
A crucial robustness test consists in estimating the impact of a future change in the news
market on current turnout. Table 2 shows the impact of a change in newspaper competition
on turnout at the previous election (falsification test). The coefficients I obtain are all non
significant and of an order of magnitude much smaller than the ones in Table 1. This suggests
that changes in the number of newspapers are not driven by election results and brings
confidence in interpreting the coefficients of Table 1 as causal effects.
11
Results are also robust to 2-ways clustering at the county-election level and are available from the author
upon demand.
15
D.Number of Newspapers
Entry/Exit
Controls
Election-State FE
R-sq
Observations
Clusters
(1)
D.Turnout
b/se
-0.004∗∗
(0.002)
All
Yes
Yes
0.27
5228
87
(2)
D.Turnout
b/se
-0.011∗∗∗
(0.003)
Only Newspaper Entry
Yes
Yes
0.27
4114
87
(3)
D.Turnout
b/se
-0.010∗∗∗
(0.003)
Only Owner Entry
Yes
Yes
0.28
3905
87
Table 1: Impact of Competition on Turnout at Local Elections
Notes: * p<0.10, ** p<0.05, *** p<0.01. Standard errors in parentheses are clustered by county. Time period
is 1947-2008. Models are estimated in first differences. All specifications include election-state fixed effects and
demographic controls.
D.Number of Newspapers
Entry/Exit
Controls
Election-State FE
R-sq
Observations
Clusters
(1)
LD.Turnout
b/se
-0.001
(0.003)
All
Yes
Yes
0.36
3287
87
(2)
LD.Turnout
b/se
-0.002
(0.008)
Only Newspaper Entry
Yes
Yes
0.37
2495
87
(3)
LD.Turnout
b/se
0.003
(0.005)
Only Owner Entry
Yes
Yes
0.38
2481
87
Table 2: Falsification Test. Impact of Competition on Turnout at Local Elections
Notes: * p<0.10, ** p<0.05, *** p<0.01. Standard errors in parentheses are clustered by county. Time period
is 1947-2008. Models are estimated in first differences. All specifications include election-state fixed effects and
demographic controls. The dependent variable is turnout at the previous election (falsification test).”
16
4.2.1
Diagnosing Bias Using Pre-trends
17
Pre-trends are a standard diagnostic for bias in panel data models. If the relationship
between ∆NCt and ∆yCt comes only from a causal effect, ∆NCt cannot be correlated with
past values of ∆yCt . On the contrary, if the observed relationship is driven by omitted
components, ∆NCt and past values of ∆yCt may be correlated.
In Figure 1 I plot the coefficient αk from the following specification:
∆yCt =
+1
X
αk ∆NC(t−k) + δ∆xCt + ∆µrt + ∆εCt
(4)
k=−1
where yCt is the change in local turnout per eligible voters and other terms are defined
as in equation (3). The prediction that newspaper entry decreases turnout correspond at the
negative spike in the plot at k = 0. It appears that there are no significant trends before or
after the event.12
4.2.2
Interaction with Market Structure
Table 3 shows how our estimated effects vary with the extent of market competition. The
model is identical to the one in Table 1, except that the independent variables of interest
are a set of indicators for the number of newspapers in the county. Interactions with the
market structure are identified by variation in the effect of entries/exits on turnout according
to the number of newspapers in the county at election t − 1. If there are no other sources of
heterogenity in the effect of newspaper entries/exists that are correlated with the number of
newspapers prior to the event, then (under our maintained assumption) these parameters can
be taken as causal estimates of the effect of the number of competing newspapers on voter
turnout.
We find no statistically significant effect of the entry or exit of a county’s second newspaper.
On the contrary, the marginal effect of the third and the fourth newspapers are negative and
statistically significant. The entry or exit of the third newspaper has a significant negative
effect of 0.8 percentage points (columns 2 and 4). The marginal effect of the fourth newspaper
is slightly lower, with a point estimate of minus 0.7 percentage points (columns 3 and 4).
In the next section I open the black-box of news-making to understand this finding.
5
Opening the Black-Box of News-Making
There is a growing concern that the quality of the media might have decreased during the
last decades, at the exact same time when the number of media available was increasing and
the media market became more competitive. Indicative is the contemporaneous increase in
the number of television channels and the often perceived decrease in quality, exemplified by
12
With only 11 elections in the sample, it is not possible to estimate equation (4) with a k higher than 1.
18
Change in Turnout per Eligible Voter
-0.04 -0.03 -0.02 -0.01 0
0.01 0.02 0.03 0.04
-12
-6
0
+6
Years Relative to Change in Number of Newspapers
+12
Notes: The Figure shows coefficients from a regression of change in turnout per eligible voters, controlling for demographics, on a vector of leads and lags of the change in the number of newspapers (see equation (4) for details). Models
include state-election fixed effects. Error bars are +/− 1.96 standard errors. Standards errors are clustered by county.
Time period is 1947-2008.
Figure 1: Local Turnout and Newspaper Entries.
D.>=2 newspapers
(1)
D.Turnout
b/se
0.002
(0.005)
(2)
D.Turnout
b/se
-0.008∗∗
(0.004)
D.>=3 newspapers
D.>=4 newspapers
Controls
Election-State FE
R-sq
Observations
Clusters
(3)
D.Turnout
b/se
Yes
Yes
0.27
5228
87
Yes
Yes
0.27
5228
87
-0.007∗
(0.004)
Yes
Yes
0.27
5228
87
(4)
D.Turnout
b/se
0.001
(0.005)
-0.008∗∗
(0.003)
-0.007∗
(0.004)
Yes
Yes
0.27
5228
87
Table 3: Turnout Effects by Number of Newspapers
Notes: * p<0.10, ** p<0.05, *** p<0.01. Standard errors in parentheses are clustered by county. Time period
is 1947-2008. Models are estimated in first differences. All specifications include election-state fixed effects and
demographic controls.
19
the multiplication of tabloid talk shows and emergence of ”Reality” television, both in France
and in the United States.
The widely reported scandal of USA Today (one of the main daily newspaper of the US)
star reporter Jack Kelley – nominated five times for a Pulitzer Prize – who fabricated substantial portions of a number of major stories and lifted quotes from competing publications
during more than ten years, illustrates this recent perception of a decrease in news quality.
Jones (2010) refers to another – but very similar – scandal, the one of Jayson Blair, a New
York Times reporter who “had managed to find a crease in the paper’s editorial oversight and
hidden out in it like a lizard in a crack”. He documents extensively how, by the late 1990s,
local TV had all but abandoned covering politics and policy, network television news having
cut foreign bureaus and replacing experienced reporters with less experienced ones. According to him, newspapers had similarly changed as a result of a more competitive environment,
cutting the newsroom to help keep profit margins at a high level.
In this section, I open the black-box of news-making to understand why more newspaper
competition leads to a lower provision of information. I first establish the existence of a
strong ”business-stealing effect”: one more newspaper in a county reduces the circulation of
incumbent newspapers in the county by nearly 25%. Then I study how it affects newspaper
operating expenses and the number of journalists. Consistent with the business-stealing effect,
I find that more competition reduces the number of journalists of incumbent newspapers
without increasing the overall number of journalists in the county. I next show that, by
reducing the number of journalists, an increase in competition reduces the production of news
by newspapers as measured by the total number of articles by issue and the total length of each
newspaper issue. Moreover, within this production of news, more competition also reduces the
share of articles on information (and increases the share of articles on entertainment).Finally
I find that more competition leads to an increase in newspaper differentiation. Overall this
seems consistent with the predictions of the model.
5.1
5.1.1
Competition and Newspaper Profitability
Three Different Levels of Analysis
Given the fact that a number of newspapers circulate across nearby counties, there are three
possible outcomes of interest:
1. County-level outcomes: data are aggregated over newspapers at the county level.
2. Newspaper-level outcomes: data are newspaper-level data. The number of newspapers – the extent of competition on the news market – is measured as the number of
newspapers in the county in which the newspaper is headquartered.
20
3. Newspaper*County-level outcomes: data for each newspaper is disaggregated between the counties in which it circulates.
Computing newspaper*county-level and the county-level outcomes is not an issue when
the only newspapers circulating in a county are headquartered in this county and do not
circulate outside. It is more problematic when a newspaper circulates across nearby counties.
In this case I use data on geographical dispersion of circulation and, for each given newspaper,
I assign to each county in which it circulates a percentage of the value of the variable (e.g. the
number of journalists, total sales, operating expenses,...) equal to its share of the newspaper
circulation.
5.1.2
The Impact on Circulation and Readership
The entry of a newspaper on a market may have a negative impact on incumbent newspapers
if there is a “business-stealing effect”: the total circulation of the entrant newspaper exceeds
the increase in the news market total circulation. Table 4 presents regression estimates of α
from equation (3) with different dependent variables. Models are estimated using OLS with
year and county fixed effects and standard errors are clustered at the county level. I focus on
the time period 1964-2011 (one time unit represents one calendar year) since I do not have
data on the geographical dispersion of circulation before 1964.
In Table 3(a), the dependent variable yct is the total circulation in the county (aggregated
over newspapers) normalized by the number of eligible voters. I expect the sign of the coefficient to be positive or nul. The size of the coefficient is informative about the magnitude
of the “business-stealing effect” (the lower the coefficient, the higher the effect). The results
show that there is a strong business-stealing effect: the number of newspapers in the county
has no statistically significant effect on the total circulation in the county, whether or not I
include demographic controls (columns (3) and (4)).
In Table 3(b), the dependent variable is the individual circulation of each newspaper in
the county ynct . Given that I find that total circulation does not increase with a new entrant,
I expect the sign of α to be negative and statistically significant. With no controls, I find that
one more newspaper in the county reduces the circulation by eligible voters of the newspapers
in the department by approximately 2 percentage points (column 1). Including demographic
controls does not change the coefficient (column 2).
The average circulation of a newspaper in a county during this time period (1964-2011)
representing 9% of the eligible voters (Appendix Table 11), this means that one more newspaper in the county reduces the circulation of incumbent newspapers by nearly 25%. There
is an important business-stealing effect.
As I underlined above, having data on both newspaper circulation and readership from
1957 to 2011 I can compute the ratio of reported readership to circulation. I find that, on
21
(a) Panel A: Total County Circulation Per Eligible Voter
Number of Newspapers
Demographic Controls
Year FE
County FE
R-sq
Observations
Clusters
(1)
Total County Circulation
b/se
-0.005
(0.005)
No
Yes
Yes
0.66
9520
87
(2)
Total County Circulation
b/se
-0.003
(0.005)
Yes
Yes
Yes
0.71
9520
87
(b) Panel B: Newspaper Circulation Per County and Eligible Voter
Number of Newspapers
Demographic Controls
Year FE
County FE
R-sq
Observations
Clusters
(1)
Newspaper Circulation
b/se
-0.020∗∗∗
(0.003)
No
Yes
Yes
0.31
9002
87
(2)
Newspaper Circulation
b/se
-0.019∗∗∗
(0.002)
Yes
Yes
Yes
0.31
9002
87
Table 4: Impact of Competition on Newspaper Circulation
Notes: * p<0.10, ** p<0.05, *** p<0.01. Standard errors in parentheses are clustered by department. Time
period is 1964-2011. Models are estimated using OLS. All specifications include year and county fixed-effects.
In Panel A, the dependent variable yct is the total circulation in the county (aggregated over newspapers)
normalized by the number of eligible voters. In Panel B, the dependent variable ynct is the individual circulation
of each newspaper in the county normalized by the number of eligible voters.
22
average, each copy is read by 2.8 individuals (with a standard deviation of 0.6). Hence the
average entry of a newspaper reduces the readership of incumbent newspapers by eligible
voters by approximately 6 percentage points.
5.1.3
The Impact on Revenues, Expenses and the Number of Journalists
In this section I estimate the impact of the number of newspapers on operating expenses,
revenues and the number of journalists using as before OLS with newspaper and year fixed
effects. I first study the impact of the number of newspapers on the costs and revenues of
each individual newspaper (newspaper-level analysis). I next estimate the “overall” impact of
changes in competition, summing operating revenues, expenses and the number of journalists
over newspapers to obtain their aggregate values at the county level (county-level analysis).
Table 5 presents OLS regression estimates of α from equation (3) for the time period
1984-2009 with different dependent variables.
In Table 4(a) column (1) the dependent variable ynt is newspaper’s total operating revenues. Nct is the number of newspapers in the county in which the newspaper is headquartered. The result shows that operating revenues decrease with the number of newspapers in
the department. An increase in the number of newspapers on the news market by 1 leads
to a decrease in operating revenues by more than 3,000 thousand euros on average, i.e. 4%
of a standard deviation. The negative impact on operating expenses is of the same order of
magnitude (column (3)), as well as the negative impact on the number of employees (4.2% of a
standard deviation) (column (4)). The impact on sales is more important, since it represents
more than 8% of a standard deviation (column (2)).
One could argue that the difference in the number of employees may be more than compensated by the fact that journalists in a more competitive environment are more skilled or
better paid. In column (5), I look at the impact of a change in the number of newspapers on
the total payroll. The result shows that one additional newspaper decreases the payroll by
nearly 963 thousand euros, i.e. 3% of a standard deviation.
Table 4(b) presents regression estimates of α from equation (3) at the county level. Indeed,
while the results above show that an increase in competition decreases the revenues and
expenses of newspapers facing a more competitive news market, one still needs to check that
at the aggregate market level the total number of journalists or total operating expenses are
not higher after the entry of a new newspaper. The results show that overall, when variables
are aggregated at the county level, there is no statistically significant effect of a change in the
number of newspapers (columns 1 to 5).
Finally in Table 4(b) I check the consistency with the previous results by presenting
regression estimates of α from equation (3) at the newspaper/county level. In column (1),
the dependent variable ycdt is newspaper’s n operating revenues realized in county c in year
23
t. Given the results in Table 4(a), I except the coefficient to be negative and statistically
significant. The results show that there is indeed a negative effect of an increase in the
number of newspapers on newspaper revenues, sales, expenses and number of journalists.
In the next sub-section I try to understand how, through its impact on the number of
journalists, competition impacts quantitatively the production of information.
5.2
5.2.1
Newspaper Content Analysis
Evidence from the Size of Newspaper Issues
As a first evidence of how on the one hand the number of journalists, and on the other
hand the degree of competition on the news market, affect the production of information by
newspapers, I compute different indicators of the size of newspaper issues: (i) the number of
words by newspaper front page (daily data for 54 newspapers over the period 2006-2012) ; (ii)
the number of articles by newspaper issue and (iii) the total number of words by newspaper
issue (daily data for 22 newspapers over the period 2005-2012). I use two different datasets
to study the impact of the number of journalists: (i) the number of journalists data (24
newspapers over the period 2005-2011); and (ii) the total number of employees data (50
newspapers over the period 2005-2011). As I underline above, there is a strong positive
correlation between the total number of employees and the number of journalists working for
a newspaper.
Evidence from Newspaper Front Page I first study the impact of the number of journalists on the total number of words per newspaper front page. The number of words on front
pages is a good indicator of the total size of newspaper issues, since the newspapers often
print the first few paragraphs of articles, that are then continued in later pages: the higher
the number of paragraphs on the front pages, the higher the number of articles in the issue.
Figure 2 shows anecdotal evidence of the strong correlation that exists between the number
of journalists working for a newspaper, and the number of words on the front page of this
newspaper. I plot for the same date (October 2nd, 2012) the front page of four different
newspapers: two national newspapers (the New York Times and Le Monde), and two local
newspapers (L’Eveil de la Haute Loire and Ouest France). These front pages illustrate very
clearly the positive correlation: while newspapers with an important newsroom like the New
York Times (1,150 journalists) have a lot of words and paragraphs of articles then continued
in later pages on their front page, while newspapers with a very small newsroom like the
Eveil de la Haute Loire (26 articles) have very few words on their front page and only large
size photos or advertisements. Moreover, it is useful to underline that this is not driven by
differences between local and national newspaper. Ouest France, a local daily newspaper,
24
(a) Panel A: Total Newspaper Costs and Revenues
Number of Newspapers
Year FE
Newspaper FE
R-sq
Observations
N
(1)
Revenues
b/se
-3175∗∗∗
(537)
Yes
(2)
Sales
b/se
-4065∗∗∗
(1012)
Yes
(3)
Expenses
b/se
-3409∗∗∗
(533)
Yes
(4)
Employees
b/se
-21∗∗∗
(6)
Yes
(5)
Payroll
b/se
-963∗∗∗
(310)
Yes
Yes
.9693148
1028
Yes
.8871496
842
Yes
.967685
1028
Yes
.9504607
1028
Yes
.9629545
1028
(b) Panel B: Total County-Level Costs and Revenues
Number of Newspapers
Demographic Controls
Year FE
County FE
R-sq
Observations
Clusters
(1)
Revenues
b/se
-910
(994)
Yes
Yes
Yes
0.95
1043
38
(2)
Sales
b/se
-1365∗
(712)
Yes
Yes
Yes
0.92
813
38
(3)
Expenses
b/se
-1296
(908)
Yes
Yes
Yes
0.96
1042
38
(4)
Employees
b/se
-5
(9)
Yes
Yes
Yes
0.94
1044
38
(5)
Payroll
b/se
-582
(600)
Yes
Yes
Yes
0.95
1043
38
(c) Panel C: Newspaper∗County Costs and Revenues
Number of Newspapers
Demographic Controls
Year FE
County FE
R-sq
Observations
Clusters
(1)
Revenues
b/se
-2501∗∗
(1063)
Yes
Yes
Yes
.7045631
2978
72
(2)
Sales
b/se
-2266∗∗∗
(583)
Yes
Yes
Yes
.6683123
2380
71
(3)
Expenses
b/se
-2542∗∗
(1020)
Yes
Yes
Yes
.7246147
2976
72
(4)
Employees
b/se
-18∗∗
(8)
Yes
Yes
Yes
.6834008
2981
72
(5)
Payroll
b/se
-997∗
(516)
Yes
Yes
Yes
.6638443
2981
72
Table 5: Impact of Competition on Newspaper Costs and Revenues
Notes: * p<0.10, ** p<0.05, *** p<0.01. Time period is 1984-2009. Models are estimated using OLS. All
variables (excepted the number of employees) are in (constant 2009) thousand euros. In Panel A, the dependent
variables are values for newspapers. Standard errors in parentheses are robust and specifications include year
and newspaper fixed-effects. In Panel B, dependent variables are values for counties. In Panel C, dependent
variables are values for newspapers by counties. In Panel B and C, standard errors in parentheses are clustered
by county and specifications include county-level demographic controls, year and county fixed-effects.
25
with a larger newsroom than Le Monde (561 against 285 journalists) has also more words on
its front page.
From an empirical point of view, the main advantage of using front pages – even if they
do not offer as much information as the entire content of news issues – is that I have more
data available for front pages, with data for 51 newspapers, i.e. mainly all the French local
daily newspapers.
In Figure 3, I plot for each newspaper the average annual total number of words on newspaper front pages compared with the newspaper’s number of journalists (3(a)) and the newspaper’s total number of employees (3(b)). There is much more data points when considering the
total number of employees since I have more data available for this variable than for the number of journalists. I find a positive correlation between the number of journalists/employees
and the total number of words on the frontpage: the more journalists/employees working for
a newspaper, the more words on the front page. Second, by plotting separately the correlation for newspapers in a county with a monopoly (“blue Plus” symbols) and newspapers in a
county with competition (“red dots” symbols), I find that this positive correlation is driven
in part by monopolistic newspapers which have much more journalists (as underlined above)
and produce a higher number of articles per issue.
I perform a regression analysis below to confirm this finding.
Evidence from the Size of Newspaper Issues
26
CMYK
Nxxx,2012-10-01,A,001,Bs-4C,E3
Late Edition
Today, partly sunny, a warmer afternoon, high 72. Tonight, increasing clouds, low 61. Tomorrow, mostly cloudy, a couple of showers, high
73. Weather map is on Page A24.
VOL. CLXII . . No. 55,911
© 2012 The New York Times
$2.50
NEW YORK, MONDAY, OCTOBER 1, 2012
Seeking Return PAYROLL TAX RISE
Of Art, Turkey FOR 160 MILLION
Jolts Museums
MISTAKEN FAITH
IN SECURITY SEEN
AT LIBYA MISSION
IS LIKELY IN 2013
Complaints by Louvre
and Met Over Tactics BREAK IS SET TO EXPIRE
By DAN BILEFSKY
ISTANBUL — An aggressive
campaign by Turkey to reclaim
antiquities it says were looted
has led in recent months to the
return of an ancient sphinx and
many golden treasures from the
region’s rich past. But it has also
drawn condemnation from some
of the world’s largest museums,
which call the campaign cultural
blackmail.
In their latest salvo, Turkish officials this summer filed a criminal complaint in the Turkish
court system seeking an investigation into what they say was the
illegal excavation of 18 objects
that are now in the Metropolitan
Museum of Art’s Norbert Schimmel collection.
Last year, Turkish officials recalled, Turkey’s director-general
of cultural heritage and museums, Murat Suslu, presented
Met officials with a stunning ultimatum: prove the provenance of
ancient figurines and golden
bowls in the collection, or Turkey
could halt lending treasures. Turkey says that threat has now
gone into effect.
“We know 100 percent that
these objects at the Met are from
Anatolia,” the Turkish region
known for its ancient ruins, Mr.
Suslu, an archaeologist, said in
an interview. “We only want back
what is rightfully ours.”
Turkey’s efforts have spurred
an international debate about
who owns antiquities after centuries of shifting borders. Museums
like the Met, the Getty, the
Louvre and the Pergamon in Berlin say their mission to display
global art treasures is under
siege from Turkey’s tactics.
Museum directors say the repatriation drive seeks to alter accepted practices, like a widely
Continued on Page A3
JIM YOUNG/REUTERS
Ryder Cup Snatched
Martin Kaymer capped Europe’s last-day rally. Page D1.
BEFORE BENGHAZI RAID
Neither Party Supports
Extending a Measure
Into Next Year
Response to June Bomb
Raised Confidence in
Local Guards
By ANNIE LOWREY
This article is by Eric Schmitt,
David D. Kirkpatrick and Suliman Ali Zway.
WASHINGTON — Regardless
of who wins the presidential election in November or what compromises Congress strikes in the
lame-duck session to keep the
economy from automatic tax increases and spending cuts, 160
million American wage earners
will probably see their tax bills
jump after Jan. 1.
That is when the temporary
payroll tax holiday ends. Its expiration means less income in families’ pocketbooks — the tax increase would be about $95 billion
in 2013 alone — at a time when
the economy is little better than it
was when the White House
reached a deal on the tax break
last year.
Independent analysts say that
the expiration of the tax cut could
shave as much as a percentage
point off economic output in 2013,
and cost the economy as many as
one million jobs. That is because
the typical American family had
$1,000 in additional income from
the lower tax.
But there is still little desire to
make an extension part of the negotiations that are under way to
avert the huge tax increases and
across-the-board spending cuts,
known as the fiscal cliff, that will
start in January without a deal.
For example, without any action,
the Bush-era tax cuts will expire
and the military and other domestic spending programs will
be reduced.
“This has to be a temporary
tax cut,” said Timothy F. Geithner, the Treasury secretary, testifying before the Senate Budget
Committee this year and voicing
the view of many in the White
House and on Capitol Hill. “I
don’t see any reason to consider
supporting its extension.”
The White House has not
pushed for an extension. “We’ll
evaluate the question of whether
we need to extend it at the end of
the year when we’re looking at a
whole range of issues,” Jay Carney, the White House press secretary, told reporters last month.
The original point of the payroll tax holiday was to stimulate
consumer spending and aid middle-income households. But now
Congress needs the money as it
struggles with vast deficits and
believes the economy can withContinued on Page A3
TOP, BRIAN SNYDER/REUTERS; ABOVE, DAMON WINTER/THE NEW YORK TIMES
Only 2 Days Left for Cramming
As President Obama and Mitt Romney began final preparations for their debate on Wednesday, a
look at previous debates shows that altering the course of a race has been no easy task. Page A13.
‘North Dakota Nice’ Plays Well in Senate Race
By JONATHAN WEISMAN
MINOT, N.D. — Heidi Heitkamp, a Democratic Senate candidate, called Leonard Rademacher a few weeks ago looking
for his vote, but Mr. Rademacher,
a 74-year-old retiree, was feeling
ill, so Ms. Heitkamp called him
back.
“I said: ‘Heidi, save your
breath. I’m voting for you,’” Mr.
Rademacher recalled, marveling
at her personal attention. “I don’t
necessarily agree with her, but I
trust her.”
Gary Volk backed Ms. Heitkamp, a former state attorney
general, after she sat for four
hours on a slab of concrete next
to what was once his house, lis-
tening to his struggles to recover
from catastrophic flooding last
year. Larry Windus’s mind was
made up by an encounter with
her opponent, Representative
Rick Berg, a Republican, that
ended with the candidate turning
his back on him.
“He’s not very personable,”
said Mr. Windus, 55, a dishwasher at Charlie’s Main Street Cafe
here.
Senate Republicans considered
the state in their column when
Senator Kent Conrad, a veteran
Democrat, announced his retirement last year. But with shoe
leather, calibrated attacks and
likability — an intangible that
goes far in North Dakota — Ms.
Heitkamp has made this a real
fight. Though North Dakota is
Proudly Bearing Elders’ Scars,
Their Skin Says ‘Never Forget’
deeply conservative and is on no
one’s presidential map as a question mark, this race could be one
of the biggest surprises of the
2012 contests. And, like all close
races this year, it could help decide control of the Senate.
Even the National Republican
Senatorial Committee conceded
in its most recent attack ad here
that Ms. Heitkamp is making
headway. “Heidi Heitkamp: You
might like her, but on the issues
she’s wrong for North Dakota,” it
said.
The contest — the state’s first
competitive one since 1986 and
probably its nastiest in modern
history — features two very different politicians with very differContinued on Page A14
Shift by Cuomo on Gas Drilling
Prompts Both Anger and Praise
By DANNY HAKIM
By JODI RUDOREN
JERUSALEM — When Eli
Sagir showed her grandfather,
Yosef Diamant, the new tattoo on
her left forearm, he bent his head
to kiss it.
Mr. Diamant had the same tattoo, the number 157622, permanently inked on his own arm by
the Nazis at Auschwitz. Nearly 70
years later, Ms. Sagir got hers at
a hip tattoo parlor downtown after a high school trip to Poland.
The next week, her mother and
brother also had the six digits inscribed onto their forearms. This
month, her uncle followed suit.
“All my generation knows
nothing about the Holocaust,”
said Ms. Sagir, 21, who has had
the tattoo for four years. “You
talk with people and they think
it’s like the Exodus from Egypt,
ancient history. I decided to do it
to remind my generation: I want
to tell them my grandfather’s
story and the Holocaust story.”
Mr. Diamant’s descendants are
among a handful of children and
grandchildren of Auschwitz survivors here who have taken the
step of memorializing the darkest
days of history on their own bodies. With the number of survivors
here dropping to about 200,000
from 400,000 a decade ago, institutions and individuals are grappling with how best to remember
the Holocaust — so integral to Israel’s founding and identity — after those who lived it are gone.
Rite-of-passage trips to the
death camps, like the one Ms.
Sagir took, are now standard for
Continued on Page A6
WASHINGTON — An effective
response by newly trained Libyan security guards to a small
bombing outside the American
diplomatic mission in Benghazi
in June may have led United
States officials to underestimate
the security threat to personnel
there, according to counterterrorism and State Department officials, even as threat warnings
grew in the weeks before the recent attack that killed Ambassador J. Christopher Stevens and
three other Americans.
The guards’ aggressive action
in June came after the mission’s
defenses and training were
strengthened at the recommendation of a small team of Special
Forces soldiers who augmented
the mission’s security force for
several weeks in April while assessing the compound’s vulnerabilities, American officials said.
“That the local security did so
well back in June probably gave
us a false sense of security,” said
one American official who has
served in Libya, and who spoke
on condition of anonymity because the F.B.I. is investigating
the attack. “We may have fooled
ourselves.”
The presence of the Special
Forces team and the conclusions
reached about the role of the Libyan guards offer new insight into
the kind of security concerns that
American officials had before the
attack on Sept. 11.
Security at the mission has become a major issue as the Obama
administration struggles to explain what happened during the
attack, who was responsible and
how the ambassador ended up
alone. [Page A12.]
Republicans and Democrats in
recent days have demanded
more detailed explanations from
the White House and State Department on possible security
lapses. “There were warnings,”
Senator John McCain, Republican of Arizona, said on CNN’s
“State of the Union” program on
Sunday.
Just how much American and
Libyan officials misread the
threat has become even more evident as they analyze the skill
with which the mortar attack at
Continued on Page A8
URIEL SINAI FOR THE NEW YORK TIMES
At Auschwitz, Livia Rebak was branded with the number
4559. Now her grandson, Daniel Philosof, has the same tattoo.
ALBANY — A few months after Gov. Andrew M. Cuomo was
poised to approve hydraulic fracturing in several struggling New
York counties, his administration
is reversing course and starting
the regulatory process over, garnering praise from environmental groups and stirring anger
among industry executives and
upstate landowners.
Ten days ago, after nearly four
years of review by state regulators, the governor bowed to entreaties from environmentalists
to conduct another study, this one
an examination of potential impacts on public health. Neither
the governor nor other state officials have given any indication of
how long the study might take.
Then on Friday, state environmental officials said they would
restart the regulatory rule-making process, requiring them to repeat a number of formal steps, including holding a public hearing,
and almost certainly pushing a
decision into next year.
The move also means that after already receiving nearly
80,000 public comments, the state
Department of Environmental
Conservation will be soliciting
more input from New Yorkers
about hydrofracking, or fracking,
as the drilling process for natural
gas is known.
The developments have created a sense in Albany that Mr.
Cuomo is consigning fracking to
oblivion. The governor has been
influenced by the unshakable opposition from a corps of environmentalists and celebrity acContinued on Page A21
INTERNATIONAL A4-12
FASHION C8
BUSINESS DAY B1-7
SPORTSMONDAY D1-8
A.L. East Still Deadlocked
A Son Has His Say
Who Needs Models?
Window on Europe’s Crisis
Giants Come Up Short
Bo Guagua, whose parents have been
embroiled in a high-profile political
scandal in China, issued a statement in
defense of his father, Bo Xilai. PAGE A4
In a performance titled “The Impossible
Wardrobe” in Paris, the actress Tilda
Swinton demonstrated that designer
clothes have a soul even without a body
PAGE C8
in them.
Divergent plans on pension costs from
Greece and Spain highlight Europe’s
troubles with debt and deficits. PAGE B1
A late offensive pass-interference call
pushed the Giants back, and Lawrence
Tynes missed a 54-yard field-goal attempt in the waning seconds in a 19-17
PAGE D1
loss to the Eagles.
The Yankees and the Orioles won and
remained tied for first in the division
PAGE D1
with three games left.
Smartphones Allowed in Class
Memorializing a Tragic Fire
To address a talent gap, Microsoft is
sending its engineers into classrooms to
PAGE B1
teach computer science.
The path has been cleared for a formal
memorial to the 100 people who died in a
PAGE A16
Rhode Island nightclub fire.
NEW YORK A18-24
NATIONAL A16-17
Schwarzenegger’s Story
In a memoir, the former governor recalls his careers, focusing on triumphs.
PAGE C1
A review by Janet Maslin.
Paul Krugman
A New York State assemblyman who is
an advocate for people with disabilities
plans to sue the organization that cares
for his mentally disabled son. PAGE A18
California has banned therapy involving
attempts to change the sexual orientaPAGE A16
tion of gay minors.
ARTS C1-6
EDITORIAL, OP-ED A26-27
Abuse Report Leads to Lawsuit
Ban on ‘Cure’ for Gay Minors
PAGE A27
U(D54G1D)y+&!#!=!#!?
(a) NYT: 1,150 journalists; 2,268 words
(b) Le Monde: 285 journalists; 1,263 words
Ouest-France
partout en France
et dans le monde
00183275
LUNDI 1 - MARDI 2 OCTOBRE 2012
Téléchargez votre édition au format
PDF, dès 6h du matin ,
sur votre ordinateur ou iPad
Canapé
ap é
Fauteuil
euil
me Ci
néma
& Home
Cinéma
0,80 €
Au Cœur de Votre à 10KM du Puy, en Velay
Intérieur RN
RN 8
88
8M
Montagnac
ontagnac 43370
Mardi 2 octobre 2012
www.meubles-aor.fr
w meubles-aor.fr
w.
Sidérurgie : Florange
ne croit plus au miracle
6-7
et 13-14
octobre
de 14h à 18h
Chaussures, bottines, pantoufles
Hommes - Femmes
0,60 €
ABONNEMENTS POSTÉS
3 mois ...................................49,00 €
6 mois ...................................96,00 €
1 an .....................................187,00 €
abonnement@leveil.fr
Tél. 04.71.09.32.14 - Fax. 04.71.02.94.08
ENTREPOT
DE
LANTRIAC
La maison explose
à Équeurdreville : un mort
Il était 21 h 50, lundi soir, quand le hameau de la Granchette, à Equeurdreville, a été
secoué par une explosion. Un pavillon
venait de sauter, les
murs soufflés par la
déflagration. Dans
les décombres, les
sauveteurs ont découvert le corps
d’un homme de
65 ans.
Page 6
BOISSY SA
Rédaction - Administration, 9 place Michelet B.P. 50024 - 43001 LE PUY - CEDEX
Rédaction : redaction@leveil.fr
Autres services : contact@leveil.fr
FABRICANT A LAUSSONNE
Aiguilhe : clap de fin
Ouest-France
00173050
Collection
Automne
Hiver
Justice et Liberté
N° 20721 www.ouest-france.fr Tél. 02 99 32 60 00
Directeur de la publication : François Régis Hutin
Relations abonnés : tél. 02 99 32 66 66
Brittany Ferries : les bateaux
reprennent la mer aujourd’hui
Page 7
Manche
Reuters
Ligne THT : les élus d’Europe écologie
sous les pylônes
Page 6
Les deux derniers hauts-fourneaux de Florange, en Lorraine, vont
fermer. La direction d’ArcelorMittal l’a annoncé hier. L’État a deux
mois pour trouver un repreneur. La majorité des 1 000 salariés touchés demeure sceptique.
Page 3
☛6
LE PUY-EN-VELAY
04 71 06 03 18
e
sur mesur
* Voir conditions en magasin
CRAPONNE/ARZON
04 71 04 09 03
LU
(c) Eveil: 26 journalists; 276 words
Guerre économique
Au plus fort d’une guerre économique qui n’a jamais aussi bien
porté son nom, le marché et sa
sempiternelle loi de l’offre et de
la demande s’avèrent de plus en
plus impitoyables. Arnaud Montebourg, chantre du protectionnisme
européen, affronte là, en première
ligne, la dure réalité de cette mondialisation effrénée. Qui pousse
un patron indien à éteindre les
derniers hauts-fourneaux de Lorraine, dans ces terres d’histoire de
France brutalisées par la violence
des mutations industrielles.
Un tel scénario renvoie au pari
de la réindustrialisation pris, pendant la campagne, par François
Hollande, aujourd’hui au pouvoir.
Osé et indispensable ? Oui, mais
très compliqué car il faut aussi donner du temps au temps. La volonté
Reportages
Des histoires
, des lieux,
TRIMESTRIEL
SEPTEMBRE
N°8
2012 3€
Des milliers
de soldats
de l’Ouest
sont tombés
sur le sol
belge
• Les Normands font la foire :
reportage sur des
rassemblements commerciaux
mais surtout humains
• La saga des conserves Larzul,
un succès breton de plus de
100 ans !
Sur les
traces
de nos poi
lus
Ces Europée
ns
poussés
à l’exil
par la crise
Les Norm
ands
font la foire
!
• Ils sont tombés en Belgique :
4 histoires de poilus de l’Ouest
3 € chez votre marchand de journaux
reportag
es
des hommes
Des histoires, des lieux,
des hommes
Marc Ollivier
00180376
Plus de 1600 spectateurs
pour Sinsemilia
ISOLATION
INTÉRIEURE
ET
EXTÉRIEURE
farouche du gouvernement d’empêcher les licenciements boursiers
et de taxer les entreprises coupables d’effectuer des coupes claires
dans leurs effectifs au seul motif
de satisfaire l’appétit des actionnaires ne réglera pas tout. Et en se
chargeant de trouver un repreneur
au site d’ArcelorMittal, l’État devra
en assumer la responsabilité en
cas d’échec, à l’heure où le BTP et
l’automobile sont en manque d’appétit d’acier.
Inventer des solutions ? Engager une politique industrielle globale tout en gérant, au coup par
coup, les situations d’urgence ?
Le défi est là. Gigantesque. Et s’il
demande de la lisibilité, de la visibilité et du courage, il doit favoriser l’innovation, la recherche, la
formation aussi. Tout en œuvrant
à l’augmentation de la taille des
PME françaises moins puissantes
que leurs consœurs allemandes.
Oui : les pistes du rebond existent.
À condition que toutes ces actions
et bien d’autres s’inscrivent dans
un projet collectif soutenu par un
État plus stratège que coercitif. Capable d’anticiper les chocs, d’entendre les remontées du terrain.
Le tout dans un dialogue social de
qualité. Responsable, constructif.
Pas dogmatique.
20364 - 1909
- 03,00
E
+10% REMISE
Pour bouquet de travaux PAGES jusqu’au 13 OCTOBRE
Renseignez-vous : www.gouv.fr
affirme, à raison, qu’elle a hérité
d’une situation avec des plans cachés sous le tapis par l’ancienne
équipe au pouvoir. Une certitude :
le débat ne devrait pas se réduire à
ces querelles politico-politiciennes
au moment où des femmes et des
hommes se battent, au quotidien,
pour sauver leurs emplois et assurer un avenir à leur famille.
En Sports
E - 3€
MAGAZIN
:< HIMKNG=YUXUU[:?b@j@
k@t@a>
Les mauvais plans de l’automne social
Ils tombent comme les feuilles
mortes. Les uns après les autres.
Jour après jour. C’était attendu,
certes. Mais les plans sociaux en
cascade minent durement cette
rentrée sociale. Et le bras de fer
engagé par Arnaud Montebourg,
le ministre du Redressement productif, avec la direction d’ArcelorMittal, le géant planétaire de l’acier,
n’est qu’un symbole de la mondialisation galopante qui, telle une
pieuvre, enserre dans ses tentacules bien des destins.
Les innombrables fermetures et
restructurations annoncées chez
Petroplus, Doux, PSA PeugeotCitroën, Sanofi, Fralib… sont les
arbres de haute tige qui cachent
une forêt dense. Faite de situations
tendues, tout aussi terribles chez
beaucoup de sous-traitants qui
disparaissent en silence. Comme
du reste dans ces petites entreprises familiales, souvent oubliées
des politiques, et en proie aussi à
de graves difficultés.
Que faire pour stopper l’avalanche et la forte poussée du chômage qui en résulte ? La droite a
beau jeu, aujourd’hui, de se mettre
en travers… du gouvernement en
raillant ou son « inaction » ou son
« incompétence » face aux désastres enclenchés. Quand la gauche
CREDIT
D’IMPÔT 18%
*
PORTES
FENÊTRES
VOLETS
BOIS - ALU
PVC - MIXTE
ESCALIERS
Page 4
par Pierre Cavret
AFP
Brice Robin (photo), procureur de Montpellier, a fait état
de « très fortes suspicions »
dans l’enquête pour paris illicites et match présumé truqué. Sont notamment visés,
neuf joueurs ou ex-joueurs du
club de hand de la ville.
Thierry Creux
C’est parfois le seul lien du
malade avec l’extérieur. Il faut
compter entre 100 et 150 €
par mois pour louer un téléviseur dans un établissement hospitalier. Un coût élevé, alors que les prestations
payantes se multiplient.
EPA
Handball : les soupçons se précisent
Les Géorgiens ont voté hier
pour élire leurs députés. La
complexité du scrutin retardant l’annonce définitive des
résultats, majorité sortante
(ici, le président Saakachvili)
et opposition revendiquaient,
en soirée, la victoire.
Page 2
sas
Page 8
À l’hôpital, la télé alourdit la facture
Commentaire
Clap de fin sur le 1050e anniversaire
de la chapelle
Saint-Michel d’Aiguilhe.
☛ 20-21
16e saison de Villes en scène :
79 spectacles, des Pieux à Mortain
Un changement politique en Géorgie ?
Désormais
également
sur iPad enrichi
de photos et
de vidéos
boutique.ouestfrance.fr
(d) Ouest France: 561 journalists; 1,367 words
Figure 2: Anecdotal Evidence: Newspaper Front Pages and Number of Journalists
Notes: October 2nd, 2012 front pages of four different newspapers: the New York Times (Figure 2(a)), Le
Monde (Figure 2(b)), L’Eveil de la Haute Loire (Figure 2(c)) and Ouest France Figure 2(d)).
27
28
Average Number of Words on the Front Page
200
400
600
800
1000
400
Number of Journalists
Number of Words on the Front Page - Monopoly
Number of Words on the Front Page - Competition
200
600
1000
Total Number of Employees
1500
Number of Words on the Front Page - Monopoly
Number of Words on the Front Page - Competition
500
2000
(b) Number of Words on Front Pages vs. Total Number of Employees
0
Notes: These Figures show the correlation between the average annual total number of words on newspaper front pages per newspaper and the newspaper’s
number of journalists (3(a)) or the newspaper’s total number of employees (3(b)). “Blue Plus” symbols are used for newspapers in counties with monopoly. “Red
dots” symbols are used for newspapers in counties with competition. Time period is 2006-2011.
Figure 3: Newspaper Front Pages and Number Journalists/Employees
(a) Number of Words on Front Pages vs. Number of Journalists
0
Average Number of Words on the Front Page
200
400
600
800
1000
In Figure 4 (respectively Figure 5), I plot the total number of articles (respectively of
words) by newspaper issue compared with the newspaper’s number of journalists or the total
number of employees. As for the number of words on front pages, it appears clearly is that
there is a positive correlation between the number of journalists/employees and the number of
articles (of words) produced by a newspaper: the more journalists working for a newspaper,
the more articles a newspaper produces everyday. Second, this positive correlation is driven
in part by monopolistic newspapers which have more journalists (as underlined above) and
produce a higher number of articles per issue.
Table 6 presents regression analysis of the impact of newspaper competition on the size
of the newspaper issues. Let c index counties, t index date (one time unit now represents one
day), y index year and n index newspapers. I assume that:
#articlesnct = α1 + α2 Ncy + α3 #journalistsny + µyear + γdayof week + εnct
(5)
where µyear is a year fixed effect and γdayof week is a day of the week fixed effet. Ncy is as
before the number of newspapers in the county in which the newspaper is headquartered in
year y, #journalistsny is the number of journalists/emplyees working for newspaper n in year
y, and εnct is a newspaper-county-day shock. #articlesnct is alternatively the total number of
words on newspaper N ’s front pages at date t, the total number of articles in the newspaper
issue and the total number of words in the newspaper issue. To adjust standard errors for
possible dependence in residuals, I use 2-ways clustering and cluster my standard errors on
newspaper and on month simultaneously.
Table 5(a) column (1) shows the impact of the number of newspapers on the news market
on the size of newspaper issues (the total number of articles). This effect is negative and
statistically significant. An increase by 1 in the number of newspapers on the news market
decreases the average number of articles in a newspaper issue by 151, i.e. one half of a standard
deviation. Column (2) shows the impact of the number of employees on the number of articles.
This impact is positive and statistically significant: a one-standard deviation increase in the
number of employees by 0.66 standard deviation. Finally in column (3) I introduce at the
same time the number of newspapers and the number of employees. Doing so, the effect of the
number of newspapers turns not to be statistically significant. On the contrary, it increases
the coefficient measuring the effect of the number of employees: a one-standard deviation
increase in the number of employees by 0.71 standard deviation. This is consistent with the
intuition that the negative impact of competition on news provision acts mainly through the
number of journalists channel.
In Table 5(b) I perform a similar analysis, except that the dependent variable of interest
is the total number of words per issue. The results are similar to the ones I obtain when
considering the number of articles. When controlling for the extent of competition, I find
29
30
400
Number of Journalists
600
1000
Total Number of Employees
1500
Number of Articles per Issue -- Monopoly
Number of Articles per Issue --Competition
500
2000
(b) Number of Articles per Issue vs. Total Number of Employees
0
Notes: These Figures show the correlation between the average annual total number of articles per newspaper issue and the newspaper’s number of journalists
(4(a)) or the newspaper’s total number of employees (4(b)). “Blue Plus” symbols are used for newspapers in counties with monopoly. “Red dots” symbols are
used for newspapers in counties with competition. Time period is 2005-2011.
Figure 4: Newspaper of Articles per Issue and Number Journalists/Employees
(a) Number of Articles per Issue vs. Number of Journalists
Number of Articles per Issue -- Monopoly
Number of Articles per Issue -- Competition
200
Average Number of Articles per Issue
500
1000
1500
0
0
Average Number of Articles per Issue
500
1000
1500
0
31
400
Number of Journalists
Number of Words per Issue -- Monopoly
Number of Words per Issue -- Competition
200
600
0
1000
Total Number of Employees
1500
Number of Words per Issue -- Monopoly
Number of Words per Issue --Competition
500
2000
Notes: These Figures show the correlation between the average annual total number of words per newspaper issue and the newspaper’s number of journalists
(5(a)) or the newspaper’s total number of employees (5(b)). “Blue Plus” symbols are used for newspapers in counties with monopoly. “Red dots” symbols are
used for newspapers in counties with competition. Time period is 2005-2011.
Figure 5: Newspaper of Words per Issue and Number Journalists/Employees
(b) Number of Words per Issue vs. Total Number of Employees
Average Number of Words per Issue
100000 200000 300000 400000
0
(a) Number of Words per Issue vs. Number of Journalists
0
Average Number of Words per Issue
100000
200000
300000
0
that a one-standard deviation increase in the number of employees by 0.75 standard deviation.
This effect is statistically and economically significant.
Finally in Table 5(c) I perform the same analysis but with the number of words per front
pages as the dependent variable. The results are going in the same direction and of the
same order of magnitude: an increase by 1 in the number of newspapers on the news market
decreases the average number of words on the front page by 63, i.e. nearly one third of a
standard deviation (column 1). However, these results are less statistically significant. In
particular, when I introduce at the same the number of newspapers on the market and the
number of employees I obtain non statistically significative coefficients – although of the good
sign – for both independent variables. This was expected: indeed different newspapers can
choose different strategies as to their front pages, e.g. more or less pictures or pictures of
different sizes, front pages are learning to us less than the entire content of news issues.
5.2.2
Distribution of Articles across Topics
In order to study the distribution of articles by topic, I use the information provided by
the website Lexis-Nexis. When I retrieve the entire newspaper issues, I also retreive all the
metadata associated with each newspaper articles, and in particular their title, topic and
subject. This metadata is defined into more details in the Data Appendix.
Combining information from the title, topic and subject, I determine for each article its
category. I create 13 different categories: agriculture, culture, economics, education, environnement, health, international, leisure activities, movies, “news in brief”, politics, religion and
sports. I define the share of articles on “information” as the number of articles on agriculture,
economics, education, environnement, international or politics, divided by the total number
of articles classified by topics. Symmetrically, I define the share of articles on “entertainment”
the number of articles on culture, health, leisure activities, movies, “news in brief”, religion
or sports, divided by the total number of articles classified by topics
Graphical Evidence In Figure 6 I plot for each newspaper the annual average share of
articles on information compare with the newspaper number of journalists/employees. I find
a positive correlation between the number of journalists/employees and the share of articles
on information. On the contrary (and by symmetry), when I plot in Figure 7 the share of
articles on entertainment compared to the number of journalists/employees, I find a negative
correlation.
In Appendix Figure 8 (respectively Figure 9), I plot the same figure but with the share
of words in articles on information (respectively on entertainment) rather than just the share
of articles, and find a similar picture. The picture is also similar when I consider a specific
example of information – the share of articles on politics in Appendix Figure 10 – or a specific
32
(a) Panel A: Number of Articles per Newspaper Issue
Number of Newspaper
(1)
# Articles
b/se
-151.481∗∗∗
(43.381)
(2)
# Articles
b/se
0.503∗∗∗
(0.150)
Yes
Yes
0.46
23977
21
84
Number of Employees
Year FE
Day of Week FE
R-sq
Observations
Clusters Newspapers
Clusters Month
Yes
Yes
0.18
24104
22
84
(3)
# Articles
b/se
35.252
(47.913)
0.546∗∗∗
(0.165)
Yes
Yes
0.46
23977
21
84
(b) Panel B: Number of Words per Newspaper Issue
Number of Newspaper
(1)
# Words
b/se
-49325.044∗∗∗
(9245.945)
Number of Employees
Year FE
Day of Week FE
R-sq
Observations
Clusters Newspapers
Clusters Month
Yes
Yes
0.26
26709
22
84
(2)
# Words
b/se
138.851∗∗∗
(33.401)
Yes
Yes
0.50
24625
21
84
(3)
# Words
b/se
10512.210
(11086.743)
151.773∗∗∗
(37.661)
Yes
Yes
0.50
24625
21
84
(c) Panel C: Number of Words per Newspaper Front Page
Number of Newspaper
(1)
# Words Front Page
b/se
-63.363∗∗∗
(22.718)
Number of Employees
Year FE
Day of Week FE
R-sq
Observations
Clusters Newspapers
Clusters Month
Yes
Yes
0.09
80728
51
62
(2)
# Words Front Page
b/se
0.170∗
(0.092)
Yes
Yes
0.10
67216
41
62
(3)
# Words Front Page
b/se
-39.022
(31.236)
0.133
(0.100)
Yes
Yes
0.11
67216
41
62
Table 6: Size of Newspaper Issues, Newspaper Competition and the Number of Employees
Notes: * p<0.10, ** p<0.05, *** p<0.01. Standard errors in parentheses are clustered by newspaper and
month. Time period is 2005-2011 for the number of articles and the number of words per newspaper issue
(Tables 5(a) and 5(b)) . Time period is 2006-2011 for33the number of words on newspaper front pages (Table
5(c)). Models are estimated using OLS. All specifications include year and and day-of-the-week fixed effects.
34
Average Share of Information Articles per Issue
%
40
45
50
35
30
25
400
Number of Journalists
Share of Articles on Information
200
600
500
Share of Articles on Information
1000
1500
Total Number of Employees
2000
(b) Share of Information Articles per Issue vs. Total Number of Employees
0
Notes: These figures show the correlation between the average annual share of articles on “information” per newspaper issue and the newspaper’s number of
journalists (Figure 6(a)) or the newspaper’s total number of employees (Figure 6(b)). The share of articles on “information” is defined as the number of articles on
agriculture, economics, education, environnement, international or politics, divided by the total number of articles classified by topics. Time period is 2005-2011.
Figure 6: Share of Information Articles per Issue and Number Journalists/Employees
(a) Share of Information Articles per Issue vs. Number of Journalists
0
Average Share of Information Articles per Issue
%
20
30
40
50
35
Average Share Entertainment Articles per Issue
%
65
70
75
60
55
50
400
Number of Journalists
Share of Articles on Entertainment
200
600
1000
1500
Total Number of Employees
Share of Articles on Entertainment
500
2000
(b) Share of Entertainment Articles per Issue vs. Total Number of Employees
0
Notes: These figures show the correlation between the average annual share of articles on “entertainment” per newspaper issue and the newspaper’s number of
journalists (Figure 7(a)) or the newspaper’s total number of employees (Figure 7(b)). The share of articles on “entertainment” is defined as the number of articles
on movies, culture, leisure activities, sports, “news in brief”, religion or health, divided by the total number of articles classified by topics. Time period is
2005-2011.
Figure 7: Share of Entertainment Articles per Issue and Number Journalists/Employees
(a) Share of Entertainment Articles per Issue vs. Number of Journalists
0
Average Share of Entertainment Articles per Issue
%
50
60
70
80
90
example of entertainment – the share of articles on sports in Appendix Figure 11. In the next
part I run a regression analysis to confirm these findings.
Regression analysis
Table 7 presents OLS regression of α2 (number of newspapers) and α3
(number of employees) from equation 5. In columns (1) to (3), the dependent variable is the
share of articles on information. I find that an increase in one in the number of newspapers
decreases the share of articles on information by 1,6 percentage points, i.e. 0.1 standard
deviation. The order of magnitude of the effect is higher when one considers the impact
of the number of employees: a one-standard deviation increases in the number of employees
increase the share of articles on information by 0.3 standard deviation. This effect is robust to
introducing the number of newspapers and the number of employees as independent variables
simultaneously (column 3).
In columns (4) to (6) the dependent variable is the share of articles on entertainment. I
find that this share statistically significantly increases with the number of newspapers and
decreases with the number of employees.
5.2.3
Distribution of Articles across Topics
Finally Table 8 presents OLS regression of α2 (number of newspapers) and α3 (number of
employees) from equation 5 when the dependent variable is newspaper specialization, as measured by the Herfindahl index described above. In columns (1) to (3) I present results of
the estimation when the Herfindahl index is computed on a daily basis, and in columns (4)
to (6) when it is computed on a weekly basis (specifications in columns (4) to (6) do not
include day-of-the-week fixed effects). The results are robust to both specification. It appears
clearly that more competition leads to more newspaper differentiation: an increase by one in
the number of newspapers in the news market leads to an increase in the Herfindahl index
by 0.05, which corresponds to a 0.4 standard deviation increase in the index of newspaper
specialization (column 1). Moroever, this is robust to controlling for the number of employees
working in the newspaper (columns (3) and (6)).
6
Conclusion
This paper questions the common wisdom whereby more competition in the media industry
leads necessarily to more information. Combining a voting model with a multidimensional
differentiation competition model I find that as long as information is costly to produce and
newspaper buyers differ little in their taste for information then, compared to a monopoly,
competition decreases the amount of information produced by newspapers and, through this
channel, decreases turnout. I confirm these predictions empirically using a new panel of French
36
37
Yes
Yes
0.06
23744
84
0.007∗∗∗
(0.000)
Yes
Yes
0.19
22059
84
(2)
% Information
b/se
(3)
% Information
b/se
-0.926∗∗∗
(0.198)
0.006∗∗∗
(0.000)
Yes
Yes
0.19
22059
84
Yes
Yes
0.15
23652
84
(4)
% Entertainment
b/se
3.578∗∗∗
(0.227)
-0.007∗∗∗
(0.000)
Yes
Yes
0.21
22057
84
(5)
% Entertainment
b/se
(6)
% Entertainment
b/se
1.201∗∗∗
(0.175)
-0.006∗∗∗
(0.000)
Yes
Yes
0.21
22057
84
Notes: * p<0.10, ** p<0.05, *** p<0.01. Standard errors in parentheses are clustered by month. The share of articles on “information” is defined as the number
of articles on agriculture, economics, education, environnement, international or politics, divided by the total number of articles classified by topics. The share of
articles on “entertainment” is defined as the number of articles on movies, culture, leisure activities, sports, “news in brief”, religion or health, divided by the total
number of articles classified by topics. Time period is 2005-2011. Models are estimated using OLS. All specifications include year and and day-of-the-week fixed
effects.”
Table 7: Share of Articles on Information / Entertainment, Newspaper Competition and the Number of Employees
Year FE
Day of Week FE
R-sq
Observations
Clusters
Number of Employees
Number of Newspaper
(1)
% Information
b/se
-1.626∗∗∗
(0.319)
38
(3)
Specialization
b/se
0.021∗∗∗
(0.002)
0.000
(0.000)
Yes
Yes
0.08
22107
84
Yes
No
0.13
3776
84
(4)
Specialization (week)
b/se
0.033∗∗∗
(0.002)
-0.000∗∗∗
(0.000)
Yes
No
0.05
3386
84
(5)
Specialization (week)
b/se
Table 8: Newspaper Differentiation, Newspaper Competition and the Number of Employees
Yes
Yes
0.11
24016
84
-0.000∗∗∗
(0.000)
Yes
Yes
0.05
22107
84
(2)
Specialization
b/se
(6)
Specialization (week)
b/se
0.019∗∗∗
(0.002)
0.000
(0.000)
Yes
No
0.11
3386
84
Notes: * p<0.10, ** p<0.05, *** p<0.01. Standard errors in parentheses are clustered by month. “Specialization” and “Specialization (week)” are Herfindahl
indexes of newspaper specialization, computed on a daily or a weekly basis. They can take values between 0 and 1. The Herfindahl index is equal to the sum of
the squares of the shares of the different newpaper topics in each newspaper issue: agriculture, culture, economics, education, environnement, health, international,
leisure activities, movies, “news in brief”, politics religion and sports. Time period is 2005-2011. Models are estimated using OLS. Specifications in columns (1) to
(3) include year and and day-of-the-week fixed effects. Specifications in columns (4) to (6) include year fixed effects.
Year FE
Day of Week FE
R-sq
Observations
Clusters
Number of Employees
Number of Newspaper
(1)
Specialization
b/se
0.049∗∗∗
(0.002)
local daily newspapers and turnout at local elections from 1945 to 2011. I find a negative
impact of newspaper competition on turnout at local elections. Opening the black-box of
news making to explain this finding, I show that an increase in competition leads to (i) a
decrease in the circulation of newspapers (business-stealing effect); (ii) a decrease in their
sales, revenues and expenses; and (iii) a decrease in the number of journalists by newspaper.
Through the decrease in the number of journalists, competition leads to (i) a decrease in the
production of news by newspapers as measured by the size of newspaper issues and (ii) within
these news, a decrease in the share of articles on information and an increase in the share
of articles on entertainment. Finally I find that more competition leads to an increase in
newspaper differentiation by constructing an index of newspaper specialization.
These results will benefit the political debate by questioning the traditional view that
media competition is necessarily beneficial. This view has motivated media ownership regulation. For example, the Federal Communications Commission (FCC) in the United States has
sought to diffuse ownership of media outlets among multiple firms in order to diversify the
viewpoints available to the public: “In sum, the modified broadcast ownership structure we
adopt today will serve our traditional goals of promoting competition, diversity, and localism
in broadcast services. The new rules are (...) necessary in the public interest.” (FCC, 2003).
The next step of this research agenda will consist in extending the analysis of the impact of
competition on the provision of information by the media to other media outlets. In particular
it would be interesting to capture to which extent the development of the Internet, which has
weakened existing media outlets in terms of audience and of ad revenues, has strengthened
this negative impact of competition on information.
39
References
Albert, P. (1989): Tirages des journaux sous la IVe République: août 1944-juin 1958,
Cahiers de l’Institut Fran{ç}ais de Presse. Institut Français de Presse et des Sciences de
l’Information.
(2004): La presse française. La Documentation Française, Paris.
Angelucci, C. (2012): “Information Acquisition, Polarization, and Persuasion,” .
Angelucci, C., J. Cagé, and R. De Nijs (2012): “The Ad Distortion Effect on Subscription Price Discrimination by Newspapers,” .
Arnold, D. R. (2002): “The Press and Political Accountability,” .
Bach, L. (2011): “Should Politicians be Forbidden Multiple Office-Holding? Evidence from
France,” .
Banerjee, A. V., S. Kumar, R. Pande, and F. Su (2010): “Do Informed Voters Make
Better Choices? Experimental Evidence from Urban India,” .
Berry, S., and A. Pakes (2007): “The Pure Characteristics Demand Model,” International
Economic Review, 48(4), 1193–1225.
Berry, S. T., and J. Waldfogel (2001): “Do Mergers Increase Product Variety? Evidence
from Radio Broadcasting,” The Quarterly Journal of Economics, 116(3).
(2010): “Quality and market size,” Journal of Industrial Economics, 58(1), 1–31.
Besley, T., and A. Prat (2006): “Handcuffs for the Grabbing Hand? Media Capture and
Government Accountability,” American Economic Review, 96(3), 720–736.
Della Vigna, S., R. Enikolopov, V. Mironova, M. Petrova, and E. Zhuravskaya
(2010): “Cross-Border Effects of Foreign Media: Serbian Radio and Nationalism in Croatia,” .
Della Vigna, S., and A. Kennedy (2011): “Does Media Concentration Lead to Biased
Coverage? Evidence from Movie Reviews,” .
Derieux, E., and J. Texier (1974): La presse quotidienne en France. Armand Colin, Paris.
Di Tella, R., and I. Franceschelli (2009): “Government Advertising and Media Coverage of Corruption Scandals,” .
Durante, R., and B. Knight (2009): “Partisan Control, Media Bias and Viewer Responses:
Evidence from Berlusconi’s Italy,” .
40
Enikolopov, R., M. Petrova, and E. Zhuravskaya (2011): “Media and Political Persuasion: Evidence from Russia,” American Economic Review.
Eveno, P. (2008): Les médias sont-ils sous influence ? Larousse, Paris.
ΩÉveno and Des travaux historiques et scientifiques
Éveno, P., and F. C. Des travaux historiques et scientifiques (2003): L’argent de la
presse fran{ç}aise des ann{é}es 1820 {à} nos jours, CTHS histoire. {É}ditions du CTHS.
Fan, Y. (2011): “Ownership Consolidation and Product Characteristics: A Study of the U.S.
Daily Newspaper Market,” .
Feddersen, T., and A. Sandroni (2006a): “A Theory of Participation in Elections,”
American Economic Review, 96(4), 1271–1282.
Feddersen, T. J., and W. Pesendorfer (1996): “The Swing Voter’s Curse,” American
Economic Review, 86(3), 408–424.
Feddersen, T. J., and A. Sandroni (2006b): “Ethical Voters and Costly Information
Acquisition,” Quarterly Journal of Political Science, 1(3), 187–311.
Gabszewicz, J. J., D. Laussel, and N. Sonnac (2002): “Press Advertising and the
Political Differentiation of Newspapers,” Journal of Public Economic Theory, 4(3), 317–
334.
Gentzkow, M. (2006): “Television and Voter Turnout,” Quarterly Journal of Economics,
121(3), 931–972.
Gentzkow, M., and E. Kamenica (2012): “Competition in Persuasion,” .
Gentzkow, M., and J. M. Shapiro (2006): “Media Bias and Reputation,” Journal of
Political Economy, 114(2), 280–316.
(2010): “What Drives Media Slant? Evidence from US Daily Newspapers,” Econometrica, 78(1).
Gentzkow, M., J. M. Shapiro, and M. Sinkinson (2011a): “Competition and Ideological
Diversity: Historical Evidence from US Newspapers,” .
Gentzkow, M., J. M. Shapiro, and M. Sinkinson (2011b): “The Effect of Newspaper
Entry and Exit on Electoral Politics,” American Economic Review.
George, L. (2007): “What’s fit to print: The effect of ownership concentration on product
variety in daily newspaper markets,” Information Economics and Policy, 19(3-4), 285–303.
41
George, L., and F. Oberholzer-Gee (2011): “Diversity in Local Television News,” .
Groseclose, T., and J. Milyo (2005): “A Measure of Media Bias,” Quarterly Journal of
Economics, 120(4), 1191–1237.
Guillauma, Y. (1988): La presse en France. La Découverte, Paris.
(1995): La presse politique et d’information générale de 1944 à 1958. Inventaire de
titres. Paris.
Herriot, J. G., and C. H. Reinsch (1973): “Algorithm 472: procedures for natural spline
interpolation [E1],” Commun. ACM, 16(12), 763–768.
INSEE (1947): Premiers résultats du recensement général de la population effectué le 10 mars
1946. Imprimerie Nationale, Paris.
(1958): Recensement général de la population de mai 1954. Résultats du sondage au
1/20ème: Population. Imprimerie Nationale, Paris.
Jones, A. S. (2010): Losing the News. The Uncertain Future of the News that Feeds Democracy. Oxford University Press.
Kayser, J. (1963): Le quotidien français. Armand Colin, Paris.
Le Floch, P., and N. Sonnac (2000): Economie de la presse. La découve, Paris.
Martin, L. (2005): La presse écrite en France au XXe siècle. Le livre de poche, Paris.
Mullainathan, S., and A. Shleifer (2005): “The Market for News,” American Economic
Review, 188(2), 355–408.
Nosko, C. (2011): “Competition and Quality Choice in the CPU Market,” .
Oberholzer-Gee, F., and J. Waldfogel (2009): “Media Markets and Localism: Does Local News en Espanol Boost Hispanic Voter Turnout?,” American Economic Review, 99(5),
2120–2128.
Prat, A., and D. Stromberg (2011): “The Political Economy of Mass Media,” .
Schulhofer-Wohl, S., and M. Garrido (2009): “Do Newspapers Matter? Short-run
and Long-run Evidence from the Closure of The Cincinnati Post,” Working Paper 14817,
National Bureau of Economic Research.
Snyder, J., and D. Stromberg (2010): “Press Coverage and Political Accountability,”
Journal of Political Economy, 188(2), 355–408.
42
Song, M. (2007): “Measuring Consumer Welfare in the CPU Market: An Application of
the Pure-Characteristics Demand Model,” The RAND Journal of Economics, 38(2), pp.
429–446.
Stromberg, D. (2004a): “Mass Media Competition, Political Competition, and Public Policy,” Review of Economic Studies, 71(1), 265–284.
(2004b): “Radio’s Impact on Public Spending,” Quarterly Journal of Economics,
119(1).
43
A
Data Appendix
A.1
A.1.1
Newspaper Data
Number of Newspapers
To determine for each year between 1945 and 2011 the number of newspapers present in each
French department, I use various sources of information that I digitize.
For the 1945-1958 period, I use as a first source of information Guillauma (1995) who lists
all the political and general information newspapers that have been published in France over
the period. I extract from this list all the local daily newspapers. I then check the consistency
of Guillauma data by using three other sources. First, the ”Cahiers de L’Institut Français de
Presse”, a standard publication from an important French institute of press studies. Second,
data from Ministry of Information reports on the state of French newspapers, that I collect in
the French national archives. Third, the Annuaires de la Presse et de la Publicité, an annual
directory of French newspapers. Newspaper directories are standard sources for historical
research on French newspapers, but have never been digitized before. They originated as a
guide to potential advertisers and were intended to be complete. However, I found a number of
mistakes in these directories. I thus always check their consistency by using at least another
data source. Particularly, newspapers often ”disappear” from the directories before their
actual exit of the market. This could have possibly biased the analysis which relies on the
timing of the changes in newspaper competition.
For the 1959-2006 period, I use the Annuaires de la Presse et de la Publicité as the first
source of information. For the 2007-2011 period, I use a more recent directory of newspapers
(”Tarif média. La première source d’information sur les médias”). I check the consistency
of this data by using circulation data (see below). Kayser (1963), Derieux and Texier (1974),
Guillauma (1988), Le Floch and Sonnac (2000), Albert (2004), Martin (2005) and Eveno
(2008)have also been useful sources of information on French local daily newspapers.
Newspaper Owners To determine the identity of newspaper owners, I use (i) studies by
historians, especially Derieux and Texier (1974); (ii) the archives of Le journaliste (a quarterly
periodical published by the ”Syndicat National des Journalistes” (SNJ) – National Union
of Journalists); (iii) the archives of the INA (Institut National de l’Audiovisuel – National
Audiovisual Institute); (iv) the archives of the newspaper Le Monde; and (v) information
provided on the website of the Graduate School of Journalism of Lille
13 .
As for the newspaper data, I check the consistency of our owner dataset by using for each
year and each newspaper at least two different data sources to determine the owner.
13
http://esj-lille.fr/
44
A.1.2
Newspaper Circulation
Aggregate Circulation For the time period 1944-1959, newspaper circulation data comes
first from Albert (1989) which is a standard source of historical research on this topic that
I digitize. However the data turns to be very incomplete. I complete by using archival data
from Ministry of Information reports on the state of French newspapers. I used three different
reports from the archives:
1. ”Tirage des quotidiens de province au printemps 1945”(local newspapers circulation
during the spring 1945). These tables are from a file called ”Local press, Political and
news publications”. They were made by the regional delegations of the Ministry of Information in major cities and are all from April 1945. Interestingly, there is information
on the political bias of the publications.
2. ”Tirage des quotidiens de province de 1945 à 1952” (local newspapers circulation between 1945 and 1952). These tables are from documents of the Ministry of Information.
For each city and each year there is the average circulation of all the local newspapers
published in the city. There is also information on the date of creation (or disappearance)
of each newspaper. When (for the year 1945) I have two different values for newspaper
circulation depending on data sources, I took the average of the two (however the results
are robust when always taking the maximum or always taking the minimum values).
3. ”Tirage des quotidiens de province de 1951 à 1958” (local newspapers circulation between 1951 and 1959). These tables have been established using Ministry of Information
documents with average monthly data for all local newspapers.
For the 1960-1974 period, I also use archival data: as discussed below (in the description of
newspaper profitability data) French newspapers were asked by the Ministry of Information
to report annually data on circulation, expenditures, revenues... I obtain access to these
confidential reports from which I digitize the information about circulation.
For 1975-1978, I complete circulation data by digitizing data from Proscop Media (the
Proscop Institute is a firm specialized in market research and marketing and geostrategic
consulting). The data was confidential at the time of its publication due to its high commercial
value (especially for advertising agencies) but Proscop Media reports are now available in the
French National Library.
Finally, for the 1979-2010 period, the newspaper circulation data is from the OJD, which is
the French press observatory whose aim is to certify the circulation data: http://www.ojd.com/.
Circulation Data with Geographical Dispersion
All the data described above is at the
newspaper level: it is the total circulation of a given newspaper, with no geographical disper45
sion between different departments when a newspaper circulates across nearby departments.
I complete it with data on geographical dispersion for the 1964-2010 period.
For 1964, 1968-1970, 1973, 1975-1978, 1980, 1981, 1983, 1985-1987, 1989, 1991, 1996 and
1999, data on geographical dispersion is from Proscop Media. For 1990-2010, data is from the
OJD (when available from both sources the data is always consistent). I ipolate the missing
values for the few years for which I do not have information. It gives me values consistent
with the aggregate circulation data.
A.1.3
Newspaper Readership
For the 1957-1992 period, the data on newspaper readership is from the ”Centre d’Etude des
Supports de Publicité” (CESP) which is a French interprofessionnal association gathering the
whole of the actors of the advertising market concerned with the study of the media audience
(advertisers, agencies and councils media, central merchandisings of space, advertising media
and controls). Between 1957 and 1992 the CESP published, every five years between 1957
and 1967 and annually starting from 1968 a study of French newspaper readers (“Etude sur
les lecteurs de la presse française”). The representative sample of the survey is drawn from
all French people 18 and older living in metropolitan France using electoral lists. It is a
random sample including between 250,000 and 300,000 individuals depending on the years.
The survey is done using a questionnaire whose main aim is to describe the behavior of French
readers at the time of the survey. The main information provided is about the reading of a
newspaper during the last period. At the time of the survey the data was confidential and
only available for the members of the association. However, while the CESP does not conduct
this survey anymore, I found in their office in Paris a copy of the results of each survey. I take
pictures and digitize the surveys for the following years: 1957, 1962, 1967, 1968, 1969, 1970,
1972, 1974, 1976, 1978, 1980, 1982, 1984, 1986, 1988, 1990 and 1992.
Between 1996 and 201114 , the data on newspaper readership comes from the Syndicat de
la Presse Quotidienne Régionale (SPQR – Local daily press syndicate) which conserves in
an electronic format the annual results of audience studies conducted to measure newspaper
readership. I get annual data covering the entire period expected 2005 in which no survey
was conducted. I ipolate the missing values for the years for which I do not have information.
These surveys mainly cover for each newspaper information on its aggregate readership.
However, for the period from 1996 to 2004 there is also information on readership by department for newspapers circulating across nearby departments.
14
There is a gap in the dataset for 1993-1995. To the extent of my knowledge there does not exist any survey
covering this period.
46
A.1.4
Newspaper Profitability and the Number of Journalists
Newspaper cost and revenue data come from firm survey data for the period 1984-2009.
Firm Data (1984-2009)
I compute annually for local daily newspapers between 1984 and
2009 a number of important economic indicators, namely sales, profits, value-added, operating
expenses (payroll, inputs, taxes), operating revenues (revenues from sales and revenues from
advertising), and the number of employees. I obtain data for 42 newspapers over the entire
period 1984-2009, and 5 for the sub-period 1993-2009. This data are from the
1. The Enterprise Survey (”Enquêtes Annuelles d’Entreprise” – EAE) of the INSEE (the
French national institute for statistics) which covers the period 1993-2009. It is a survey
done each year which covers firms of more than 20 employees.
2. The files constructed for the tax regime ”Bénéfice Réel Normal” (BRN) by the ”Direction Générale des Impôts” (DGI).
I identify local daily newspapers in this dataset using the registry of establishments and
enterprises called the Sirene.
Number of Journalists Data (1999-2012) The data on the number of journalists have
been provided to me by the SPQR for the period 1999-2012. I have data for 24 newspapers
with the following variables: (i) number of managers; (ii) number of journalists; (iii) number
of white-collar workers; (iv) number of blue-collar workers; and (v) total number of workers.
A.1.5
Newspapers’ Content
Front pages Newspapers’ front pages come from the SPQR website which publishes every
day the ”front pages of the day” of 54 local daily newspapers: http://www.pqr.fr/editeurs/lesunes-du-jour/. I download using all the available front pages in ”.pdf” format from their
website an automated script and for each frontages, count the number of words after having
converted ”.pdf” files in ”.txt” files using OCR softwares.
Full text
I obtain the entire daily content of each newspaper issue by using an automated
script to retrieve for each day and each newspaper issue all the articles published in the issue.
I obtain this information by downloading the information available on two different websites,
Factiva and Lexis-Nexis, which aggregate content from newspapers.
I retrieve the entire newspaper content from Factiva for 18 newspapers (beginning date
in parentheses): Berry Républicain (2010-04-01); Charente Libre (2005-05-06); Centre Presse
Aveyron (2006-09-01); Est Républicain (2008-02-27); Indépendant (2006-09-01); Maine Libre
47
(2011-03-04);Midi Libre (2006-09-01); Montagne (2010-04-01); Nouvelle République (2011-0112);Ouest France(2002-07-17); Parisien (2005-06-15); Populaire du Centre (2010-04-01);Presse
Océan (2008-10-01); Progrès (2003-10-23); République du Centre (2011-05-02); Sud Ouest
(2003-09-22); Voix du Nord (2011-02-01); Yonne Républicaine (2010-04-01).
I obtain data from Lexis-Nexis for 21 newspapers: Berry Républicain (2010-03-22); Centre
Presse Aveyron (2010-03-22); Est Républicain (2008-02-07); Havre Libre (2008-01-05); Havre
Presse (2008-01-07); Indépendant (2007-05-11); Journal Du Centre (2010-03-22); Maine Libre
(2011-09-05); Midi Libre (2006-11-01); Montagne (2010-03-22); Nouvelle République (200403-23); Ouest France(2006-04-20); Paris Normandie (2004-09-02); Parisien (2006-12-20);
Populaire du Centre (2010-03-22); Presse Océan (2010-12-08); Progrès de Fcamp (2008-01022); Sud Ouest (1994-05-07); Tégramme (2002-02-01); Voix du Nord (2009-09-14); Yonne
Républicaine (2010-03-22).
Combining results from Factiva and from Lexis-Nexis I have data on entire newspaper
issues for 24 different newspapers.
I next use the metadata associated with each article on Lexis-Nexis (title, subject, topic)
to classify articles between “information” and “entertainment”. The share of articles on
“information” is defined as the number of articles on agriculture, economics, education, environnement, international or politics, divided by the total number of articles that I am able
to classify. The share of articles on“entertainment” is defined as the number of articles on
movies, culture, leisure activities, sports, “news in brief”, religion or health, divided by the
total number of articles that I am able to classify. (By construction, the sum of both shares
is equal to 100).
Finally I use the article classification in sub-categories to construct a measure of newspaper
differentiation. This measure is simply an Herfindhal index varying between 0 – no specialization, i.e. no differentiation between newspapers that all deal with all the topics – and 1
– perfect newspaper specialization, i.e. important newspaper differentiation, each newspaper
being specialized in a given topic (e.g. music or sport). This index is equal to the sum of the
squares of the shares of the different newpaper topics in each newspaper issue: agriculture,
culture, economics, education, environnement, health, international, leisure activities, movies,
“news in brief”, politics religion and sports. I compute it both on a daily (considering each
newspaper issue separately) and or a weekly basing (summing all the issues of each newspaper
for each given week together).
48
A.2
Electoral Data
A.2.1
Local (Mayoral) Elections
Between 1947 and 2008, 11 local elections took place: in 1947, 1953, 1959, 1965, 1971, 1977,
1983, 1989, 1995, 2001 and 2008.
I digitize data for the 1947 and 1953 elections from the government archives in Paris
(available in various boxes beginning with shelf mark ”F/1cII/”). The data covers all cities
with more the 2,500 inhabitants. The names of the candidates are not available but there is
information on their political affiliation (name of the party).
For the 1959, 1965, 1971, and 1977 elections, I digitize data from the newspaper Le Monde.
This information available is only for cities with more than 9,000 inhabitants, but the names
of the candidates (on top of the party names) are available. I supplement the 1959 data for
cities under 9,000 inhabitants using data from the national archives.
For the 1983, 1989, 1995 and 2001 elections, I first use the data of the Centre de Données
Socio-Politiques (CDSP) of Science-Po Paris for all cities over 3,000 inhabitants. Candidates
names are not available in this dataset. I supplement it with data from Bach (2011) from
which I extract the candidates names.
Finally, I obtain the 2008 elections results from the Interior ministry (for all the communes
over 3,500 inhabitants and with the candidates names)
A.3
Demographic Data
I obtain commune-level demographic data from the French Census whose data is available
on the INSEE website for the censuses of 1968, 1975, 1982, 1990, and 1999, and annually since 2004.15 For the 1962 census, I obtain data from the Centre Maurice Halbwachs:
http://www.cmh.ens.fr/greco/adisp.php. I digitize the data for the 1954 census from INSEE
(1958) and the data for the 1946 census and the 1936 census from INSEE (1947).
B
French Local Juridictions & French Voting System
France is organized in six different levels of local juridictions: regions, departments, arrondissements, cantons (which can be considered as the equivalent of the US counties), ”intercommunalités” and communes, among which only four correspond to electoral circumscriptions:
regions, departments, cantons and communes.
A department is a French administrative division. There are 101 French departments.
The median land area of a department is 2,303 sq mi, which is slightly more than three-andhalf times the median land area of a county of the United States.
15
http://www.insee.fr/fr/bases-de-donnees/default.asp?page=statistiques-locales.htm
49
Number
Average pop (nb)
Average area (km2 )
Region
State
22
2,839,500
24,865
Department
County
96
650,719
5,698
Canton
Commune
3,883
16,088
141
36,570
1,722
14.88
Commune
>9,000
1,011
61,789
541
Table 9: France Local Juridictions (2008)
Sources: INSEE.
Turnout Rate at Local Elections
Observations
mean/sd
0.680
(0.09)
8126
Table 10: Descriptive Statitics: Turnout Rate at Local Elections
Notes: The Table presents the average and the standard deviation (between parentheses) of the turnout rate
at local elections in France between 1947 and 2008 (11 elections).”
I use in the article the term “county” when refering to a department, and the term “state”
when refering to a region.
The French voting system for local (mayoral) elections is the two-round list system
with proportional representation (”scrutin de liste à deux tours avec représentation proportionnelle”). The system functions as follows: if a list obtains the absolute majority in the
first round, then a number of seats equal to half of the available seats is attributed to this list.
The other seats are shared between all the other lists following the proportional representation
with the highest averages method. If no list obtains the absolute majority in the first round,
then a second round takes place. The only lists that can take part in this round are the ones
which obtained more than 10% of the recorded votes in the first round. A number of seats
equal to half of the available seats is attributed to the list which obtains most votes and the
other sears are shared between all the other lists following the proportional representation
with the highest averages method.
In Table 10 I present summary statistics of turnout at local elections in France between
1947 and 2008.
C
Descriptive Statistics
50
Total County Circulation Per Eligible Voter
Newspaper Circulation Per County and Eligible Voter
Total County Circulation
Newspaper Circulation Per County
Total Newspaper Circulation
mean/sd
0.22
(0.08)
0.09
(0.08)
80303
(70793)
31870
(39604)
216972
(474213)
Ratio of Reported Readership to Circulation
mean/sd
mean/sd
mean/sd
2.93
(1.42)
31.35
(18.63)
Newspaper Readership Per County and Eligible Voter
Newspaper Operating Revenues
97718
(79262)
53154
(49900)
96020
(77154)
692
(495)
40638
(30901)
Newspaper Sales
Newspaper Operating Expenses
Newspaper Number of Employees
Newspaper Payroll
Newspaper Number of Journalists
% Journalists in Employees
Observations
9714
8346
3312
Table 11: Descriptive Statitics: Newspaper Circulation, Cost and Revenue
Notes: All variables (excepted circulation, readership and the number of journalists and employees) are in
(constant 2009) thousand euros. Time period is 1984-2009 and variables are values for newspapers for all
variables excepted circulation, readership and the number of journalists. For circulation, time period is 19642011. For readership, time period is 1959-2011. For the number of journalists, time period is 1999-2012. The
table presents the average and the standard deviations (between parentheses) of the variables.
51
282
(152)
37
(9)
327
mean/sd
373
(225)
# Words Front Page
# Articles
mean/sd
mean/sd
423
(297)
107668
(82019)
287
(40)
# Words
Average Article Length
% Information
% Entertainment
Share of Words on Information
Share of Words on Entertainment
Newspaper Specialization
Newspaper Specialization (weekly)
Observations
82588
28541
34.8
(13.4)
66.4
(11.5)
32.7
(14.0)
68.6
(11.9)
0.17
(0.13)
0.14
(0.06)
26876
Table 12: Descriptive Statitics: Newspaper Content
Notes: Time period is 2006-2011 for newspaper front pages and 2005-2011 for the content of newspaper issues.
Variables are values for newspapers. “Newspaper specialization” and “newspaper specialization (weekly)” are
Herfindahl indexes of newspaper differentation, computed on a daily or a weekly basis. They can take values
between 0 and 1. The Herfindahl index is equal to the sum of the squares of the shares of the different
newpaper topics in each newspaper issue: agriculture, culture, economics, education, environnement, health,
international, leisure activities, movies, “news in brief”, politics religion and sports. The table presents the
average and the standard deviations (between parentheses) of the variables.
52
D
D.1
Additional Figures
Newspaper Content
53
54
Average Share of Information Words per Issue
%
40
50
60
30
20
400
Number of Journalists
Share of Words on Information
200
600
0
500
Share of Words on Information
1000
1500
Total Number of Employees
2000
(b) Share of Information Words per Issue vs. Total Number of Employees
Average Share of Information Words per Issue
%
20
30
40
50
Notes: These figures show the correlation between the average annual share of words on “information” per newspaper issue and the newspaper’s number of
journalists (Figure 8(a)) or the newspaper’s total number of employees (Figure 8(b)). The share of words on “information” is defined as the number of words on
agriculture, economics, education, environnement, international or politics, divided by the total number of words classified by topics. Time period is 2005-2011.
Figure 8: Share of Information Words per Issue and Number Journalists/Employees
(a) Share of Information Words per Issue vs. Number of Journalists
0
10
55
Average Share of Entertainment Words per Issue
%
60
70
80
50
40
400
Number of Journalists
Share of Words on Entertainment
200
600
500
Share of Words on Entertainment
1000
1500
Total Number of Employees
2000
(b) Share of Entertainment Words per Issue vs. Total Number of Employees
0
Notes: These figures show the correlation between the average annual share of words on “entertainment” per newspaper issue and the newspaper’s number of
journalists (Figure 9(a)) or the newspaper’s total number of employees (Figure 9(b)). The share of words on “entertainment” is defined as the number of words on
movies, culture, leisure activities, sports, “news in brief”, religion or health, divided by the total number of words classified by topics. Time period is 2005-2011.
Figure 9: Share of Entertainment Words per Issue and Number Journalists/Employees
(a) Share of Entertainment Words per Issue vs. Number of Journalists
0
Average Share of Entertainment Words per Issue
%
50
60
70
80
90
56
Average Share of Articles on Politics per Issue
%
10
15
5
400
Number of Journalists
Share of Articles on Politics
200
600
500
Share of Articles on Politics
1000
1500
Total Number of Employees
2000
(b) Share of Articles on Politics per Issue vs. Total Number of Employees
0
Notes: These figures show the correlation between the average annual share of articles on “politics” per newspaper issue and the newspaper’s number of
journalists (Figure 10(a)) or the newspaper’s total number of employees (Figure 10(b)). The share of articles on “politics” is defined as the number of articles on
politics divided by the total number of articles classified by topics. Time period is 2005-2011.
Figure 10: Share of Articles on Politics per Issue and Number Journalists/Employees
(a) Share of Articles on Politics per Issue vs. Number of Journalists
0
Average Share of Articles on Politics per Issue
%
5
10
15
57
Average Share of Articles on Sports per Issue
%
15
20
25
10
5
400
Number of Journalists
Share of Articles on Sport
200
600
500
Share of Articles on Sport
1000
1500
Total Number of Employees
2000
(b) Share of Articles on SPORTS per Issue vs. Total Number of Employees
0
Notes: These figures show the correlation between the average annual share of articles on “sports” per newspaper issue and the newspaper’s number of
journalists (Figure 11(a)) or the newspaper’s total number of employees (Figure 11(b)). The share of articles on “sports” is defined as the number of articles on
sports divided by the total number of articles classified by topics. Time period is 2005-2011.
Figure 11: Share of Articles on SPORTS per Issue and Number Journalists/Employees
(a) Share of Articles on SPORTS per Issue vs. Number of Journalists
0
Average Share of Articles on Sports per Issue
%
10
20
30
40
0