2.1.1. Distrust of the Internet is growing

On the contrary, the Internet is continuing to arouse a certain amount of mistrust: only one in four French people (25%) consider the information it contains to be credible, according to the media trust barometer of the La Croix newspaper. When the French get information on the Internet, they prefer newspaper sites or applications (28%) rather than social networks (18%). However, more than a third of them do not know how to clearly identify their sources of information (37%).

The study focuses on social networks: distrust here is at its highest. In this study, 66% clearly indicate that they do not trust information from social networks (such as Facebook or Twitter), even when it is published by a friend. Even more surprisingly, 45% do not trust the information, even when it is published by media.

Other studies support this view. An OpinionWay survey conducted for JIN, a public relation agency, was published in the fall of 2017, showing that despite growing consumption of these networks, Internet users do not believe everything they read on them. On a scale of 0 to 5, 74% of French people give a score of 0 to 2 to Twitter in terms of trust, and 69% of people express a similar opinion for Facebook.

Another study, the French people’s digital technology trust barometer, which was conducted by Harris Interactive for ACSEL (Association de l’économie du numérique or the Digital Economy Association) and La Poste³, and was published at the end of 2017, shows similar results. While 75% of French people say they trust the information on traditional online media sites, this figure drops to 25% for information circulating on social networks.

2.1.2. But criticism of journalists remains strong

Despite the renewed trust in traditional media, particularly in the face of social networks, the French still have a poor opinion of journalists, or more precisely of their relations with economic and political power. The media trust barometer for La Croix attests it as follows: 68% (+1 point) of French people believe that the latter are not independent of the pressures of political parties and power, and 62% (+4 points) believe that they are not independent of the pressures of money.

Even journalists regularly criticize themselves. Many essays have pointed to the abuses of journalistic practices. Recently, in a vitriolic book, La pensée en otage (Lancelin 2018), Aude Lancelin, the former deputy director of Marianne and Obs, expressed concern about, among other things, the massive investment of CAC 40 companies in the media. The book’s subtitle, S’armer intellectullement contre les médias dominants (“intellectually arming oneself against the dominant media”), sets the tone: the author criticizes the world of media, targeting both press owners and journalists, reminding the reader in particular that the deteriorating job market in media has destabilized the balance of power between writing and management.

The author tries to disassemble seven ideas that she presents as false (shareholders do not intervene; we cannot do without large private capital; criticizing the media is like attacking people; there is diversity and “media” does not exist; journalists must be neutral; newspapers are by definition democratic forces to be defended no matter what; the media cannot do much) which “prevent the public from becoming aware of the need to take up the media issue and make it a priority political issue”.

2.2.1. The measurement of an informational bias

The first task is therefore to define what differentiated content is in the economic sense of the term, in other words: in relation to an average. We could average a lot of things for example, establishing the mood of an article or newspaper, calling it optimistic or pessimistic, light or serious and so on. The list is long and could be very imaginative. Whatever the criterion that is chosen, in order to establish a qualification in any field, it is necessary to go through the content of articles and newspapers; thus, basically, the words chosen, and even the layout – which is far from impartial in the way in which information is valued – could also be taken into account. This type of analysis is now made even easier by the digitalization of content, which makes it possible to use appropriate databases that are sorted by country, newspaper, year, etc., and by software that makes it possible to sort through databases using keywords. The approach that is most often adopted by economists is that of political ideas and the ranking of a newspaper in relation to other newspapers.

Why political ideas? Because what interests the economy is economic results, such as changes in unemployment or purchasing power; and because these results are the result of the clash of individual choices, which are derived from raw or non-raw information received, beliefs, convictions and public policies, which are themselves chosen by politicians and bureaucrats driven by raw or non-raw received information, beliefs, convictions, etc.

The approach in terms of content or ideas has long been neglected by economists. Following the seminal article by Jean Gabszewicz, Didier Laussel and Nathalie Sonnac (Gabszewicz et al. 2001) published in 2001, many authors are involved in this major subject.

The researchers’ method is done in two stages: first, to measure the political bias and then to investigate its causes. The bias is highlighted by an econometric analysis of the content and frequency of articles. To do this, the research uses the processing of databases that integrate digitalized articles. Identifying the frequency of words or themes in articles, comparing it with the frequency in other media or in politicians’ speeches, is the way to analyze possible bias, whereas the cause of the bias is analyzed using a model, where the objective is to isolate the relative weight of the players involved in direct or indirect decision-making. Depending on the models, the bias is on the demand side, the supply side or a combination of both.

As Matthew Gentzkow, Jesse Shapiro and Daniel Stone pointed out in the second volume of the Handbook of Media Economics in 2016 (Gentzkow et al. 2016), economic research on press content and the origin of information biases offers two types of explanations for this bias.

2.2.2. Tests on U.S. media

Much of the research on bias concerns U.S. newspapers. This is essentially for a practical reason: in order to test the hypothesis of an informational bias, it is necessary to have large databases and computer resources. However, the digitalization of the press first took place in the U.S. press.

In 2005, (Groseclose and Milyo 2005), Tim Groseclose, a professor at George Mason University, Adam Smith, Head of the Chair, and Jeffrey Milyo a professor at the University of Missouri, were among the first to use content digitalization to measure the number of times keywords, which were chosen for their political representation, were used in certain media. They compared this number with the number of times these same words were used by elected officials in the U.S. Congress, both Republicans and Democrats. By comparing the frequency of keywords used, they ranked the media according to the ideological bias they adopted (in this case, Democrat or Republican). The basis of the test is constituted by all the speeches delivered by the elected representatives between the 103rd and 107th Congresses (between 1993 and 2002). The average frequency of keyword use among Republicans or Democrats is used as a reference base. As a result, newspapers or media are classified as neutral, or with a liberal bias in the Anglo-Saxon (i.e. Democrats) or conservative (i.e. Republicans) sense. The sample selected consists of 20 media outlets. The authors define the concept of bias by:

“the choice of a journalist to not broadcast information or to not give it its rightful place corresponding to its importance” (p. 20).

The result of their study is that there is a centrist media group with USA Today, ABC Good Morning America or CNN, a rather Democratic group with CBS, the New York Times or the Washington Post and a rather Republican group with Fox News or the Washington Times.

The analysis is based on the number of times the names of 200 think tanks are mentioned in speeches or articles, all of which are clearly positioned politically. The choice of these keywords was the subject of much discussion. As we will see, other authors taking other references result in a different positioning of some newspapers.

Riccardo Puglisi, a researcher at the London School of Economics, provides an econometric focus on the New York Times during the period 1946–1994 (Puglisi 2011). He counts the subjects covered by the newspaper. He sorts the topics according to themes qualified as Democrats (civil rights, health, labor market, etc.) and Republicans. He deduced that the New York Times has a Democratic editorial line. This positioning is particularly clear during election periods. When the incumbent President is a Republican, there is a 26% increase in papers on Democratic topics in the three months preceding the election. This increase does not occur when the outgoing President is a Democrat.

Puglisi and Snyder, also in 2011 (Puglisi and Snyder 2011), analyzed the contents of 200 U.S. newspapers covering 32 political scandals. Using a semantic keyword search, they identify newspapers with Democratic sensibilities by realizing that they do not discuss the scandals on their side and propagate articles on scandals caused by Republican politicians, and vice versa for the Republican newspapers. Sandra Garcia-Uribe (Garcia-Uribe 2016), for example, bases the measurement of information bias on the editorial choices of the covers or first pages (place given, choice of topics covered) between 2007 and 2012 for the major U.S. newspapers, and comes to similar conclusions.

In a team with two other researchers, Puglisi is continuing this research. Along with Valentino Larcinese, a professor at the London School of Economics, and James Snyder of MIT (Larcinese et al. 2011), Puglisi shows that pro-democracy newspapers tend to propagate articles on unemployment and labor market difficulties when the President is Republican, in order to highlight the latter’s poor macroeconomic performance to the public. The results are similar for the trade deficit. On the contrary, partisan behavior does not appear when we look at articles about inflation, no doubt because it is an area of politics that is less divided. The three researchers analyze the press between 1996 and 2005, and more specifically, the Los Angeles Times, owned by a family – the Chandlers – from 1884 to 2000. When Otis Chandler took over the management of the newspaper in 1960, the latter was identified as rather conservative, before then becoming favorable to the Republicans. Otis Chandler wanted to make his newspaper a credible competitor of the New York Times. This concrete case allows the authors to illustrate the weight of an owner’s influence. The particularly interesting point of this study is that is not only shows that an editorial line changes over time, but also that the reason behind it is clearly the influence of the owner. The article and the survey do not say how this change took place within the newspaper, for example if the teams of journalists changed within the political service.

In 2015, Puglisi and Snyder (Puglisi and Snyder 2015) conducted further research on the identification of ideological bias. They classified the research into three categories: research that identifies bias by comparing expressions used by newspapers and those used by politicians; research that analyzes the agenda, in other words, how information is reported in the media (subject, frequency); and lastly, research that focuses on media that support a politician or a party (endorsement). The authors propose a summary of these different approaches by constructing a model that integrates a larger number of variables and simultaneously takes into account the strategies of newspapers, pressure groups, political parties and voters.

The interesting result here is the highlighting of the heterogeneity of biases within the same newspaper that can adopt rather democratic positions on social issues and a more conservative position on economic policy, despite the global position of the newspaper on one side or the other. They conclude that the informational bias is less pronounced than in the analyses conducted by other researchers.

The identification of bias is therefore very complex. The nature of bias is equally important, since it may not be unique and may be heterogeneous within a newspaper or press group.

In 2014, John Lott and Kevin Hassett (Lott and Hassett 2014) developed an econometric analysis based on the Nexis/Lexis database from 1985 to 2004. The database contains 389 press titles. The study mainly focuses on economic articles. The results of the analysis are that the coverage of economic events is presented in a more positive way when the President is a Democrat than when he is a Republican. To do this, the authors examine how the unemployment results are presented. The difference is significant as it is 30% in favor of democrats. This bias itself would lead to a 7% improvement in public opinion regarding the results of the President’s economic policy. The result is even clearer if we take the top 10 of the press (40% instead of 30%). The study shows that economic articles play an important role in shaping public opinion and that there is an ideological bias on these subjects in the U.S. press.

2.2.3. The case of the Asian and European press

Most of the research on the concept of bias has been conducted from U.S. media. Some authors propose to extend the analysis to other countries, such as South Korea and China (Hang 2006), as well as to some European countries.

Han Yuan, from the University of Hong Kong, studies the Chinese press (Yuan 2016). The author relies on press coverage of the 18th Party Congress in November 2012 in 21 local newspapers in seven provinces, as well as in the national newspaper People’s Daily (the People’s Daily is the official media of the Chinese Communist Party. Its daily circulation is approximately three million copies⁶). While the first measures of information bias in the United States were based on the existence of a political system with two parties and the media’s preference for one of them, the same method cannot be applied in China. The author proposes to extract a matrix of words representative of the journal from the articles of each journal. This does not reveal a bias in the usual sense of the word, since all newspapers are controlled by the Communist Party, which determines the editorial line, but it does reveal significant changes that concern geographical similarities or differences, as well as the proportion of political information in relation to economic information. Concerning geographical areas, newspapers in the same region use very similar vocabulary, which is consistent with the fact that the Communist Party is organized into committees; each committee controls a specific region and therefore the newspapers published there. The author concludes that although the information is controlled by the party, there are nevertheless significant variations between newspapers.

Ralf Dewenter, University of Hamburg, Uwe Dulleck, University of Queensland in Australia, and Tobias Thomas researcher at DICE (Düsseldorf Institute for Competition Economics), (Dewenter et al. 2016), are building a media bias index by compiling 35 German media outlets (including five daily and 10 weekly newspapers) and over seven million articles from 1988 to 2012. The authors complete the computerized linguistic analysis with the results of Media Tenor International⁷. Qualifying the bias during this period is all the easier, as this period can be characterized as bi-party (CDU/CSU vs. SPD). In this recent article, a new element is introduced into the rationale. The media now play a role in people’s perception of, for example, economic performance. In other words, the media influences economic sentiment. The results show that political direction evolves over time, but one thing in common is that the media are generally harder on the ruling government than on any opposition party. The authors speak of government malice or anti-government bias.

In a 2015 article, Mark Kayser and Michael Peress (Kayser and Peress 2015) analyzed two million articles in 32 daily newspapers in 16 countries⁸. The period analyzed varies according to the newspapers (due to a problem of access to digitalized archives which allows the identification of an editorial line by keyword counting). It covers the 20-year period before 2014. The study is very complex to conduct given the diversity of languages used. Nevertheless, it shows that the press publishes more positive articles on the economy when there is a period of growth and when the newspaper has a position close to the political team in power, whereas the number of critical articles increases when there is both an economic decrease and a team in power that is not supported by the newspaper.

Content econometrics seems to show that most media adopt an editorial line that includes a choice that is favorable to an ideological movement and/or to politicians. Further research is needed. It remains to be seen whether this trend is increasing or moderating over time and whether this media bias is identified, accepted or rejected by readers.

2.2.4. The impact of newspaper owners

Recent developments in France constitute a particularly rich field of analysis. A whole stream of research is interested in the impact that a newspaper owner can have on a media outlet. This is a fundamental issue that is very often raised in France, where most of the major newspapers are privately owned and often have large companies whose main activity is not in the media sector. In the space of a few years, the media landscape has been opened up with the takeover of major media (Le Monde, Libération, L’Express, Le Parisien, etc.) by investors such as Xavier Niel, Patrick Drahi, Vincent Bolloré and Bernard Arnault, to name but a few recent and emblematic changes (De Rochegonde and Sénéjoux 2017)⁹. Economic research provides several interesting elements for reflection.

The 2003 article by Djankov et al., which has already been mentioned, launched a major debate that has been of burning relevance ever since. This debate, which now dates back almost 20 years, was recently updated by the report by Cagé and Godechot (Cagé and Godechot 2017), which has the same title as the 2003 seminal article: “Who owns the media?”. Both teams come to opposite conclusions. The former shows the significant biases associated with public media and prefers competition between private media, while the latter, concerned with owners invested in sectors other than media, offers donor-funded media far from the search for profitability (Cagé 2016)¹⁰. In the same vein, Giacomo Corneo (Corneo 2006), a professor at the University of Berlin, published an article in 2006 in which he showed a correlation between media concentration that reduces competition and the increase in information bias, using a game theory model.

Gilat Levy and Ronny Razin, researchers at the London School of Economics and Ines de Barreda (Levy et al. 2016), from Oxford University, are interested in the power of media owners in an environment where readers do not identify this power and the resulting information biases.

The authors mention a correlation neglect, in other words a difficulty for players to identify the link that may exist between different media, when they belong to the same owner. Readers consume information from several media as if it were broadcast independently, without any link between them, although there may be a correlation. This correlation is all the more difficult to achieve because in a large number of media groups, information is transferred from one to the other after being repackaged in the colors of each media. In the most frequent research, the identification and measurement of media bias was done on a media-by-media basis, independently. A more complete measure of media bias should take into account the correlation between media since an owner can integrate it into his strategy, in order to take advantage of readers’ tunnel vision about the correlation capabilities of the owner. The owner’s power increases with the number of shares he or she owns. It is therefore probably higher than measurements made in previous research. The authors conclude their article by showing that readers always benefit from the dissolution of a press group, even when all new media owners share the same ideological position.

In a 2006 article, Soontae An, Hyun Jin and Todd Simon of the University of Kansas analyzed the financial performance of U.S. media groups (An et al. 2006). The issue raised involves the possible weakening of the informative role of the media in favor of the search for financial performance. The authors distinguish two types of funders. The first type are institutional investors (banks, insurance companies, pension funds, etc.), who are the most present. In 2004, they owned 90% of the press titles of 11 out of 15 of the largest publicly traded media groups. The second ones are investors who run the media. The study of 37 local newspapers shows that media run by managers have higher profitability than media directly run by their owners. The main reason for this is the objectives of owners, who are often more ideological than they are financial, which is generally the opposite for managers. The studies included 12 journals over 13 years.

2.2.5. Pluralism and competition

Another area of research focuses on the impact of competition on the quality of media information. In other words, does the fact that there are several media outlets make for more credible media?

In Media Bias and Reputation, published in 2006 by Matthiew Gentzkow and Jesse Shapiro (Gentzkow and Shapiro 2006)¹¹, the two authors show that biases are less significant if there are several independent media in competition. Each group of readers with a type of conviction trusts a media outlet, but the number of competing media ensures a plurality of opinions. Indeed, a media outlet’s reputation also depends on the quality of the information disseminated in the past. If a media outlet alters information, it takes the risk of having it invalidated by facts and other media. Thus, competition in the supply of information is a way to reduce informational biases.

A few years later, in 2010 (Gentzkow and Shapiro 2010), the two researchers expanded their analysis by publishing a model to measure and separate the different origins of information bias. They conducted an econometric analysis of the words and expressions used by Democrat and Republican elected officials during the 2005 Congress session, using two databases (Newslibrary database and ProQuest Newsstand database). This indicator makes it possible to classify media from a database by grouping 433 newspapers (or about 74% of newspaper distribution in the USA at the time of the survey). Then, an analysis of the postal codes of newspaper subscribers makes it possible to map the distribution areas and compare the number of subscribers to the results of the electoral votes by area. The bias related to readers’ beliefs would explain most of the total bias, while the bias related to owners’ ideology would be low. This weakness, in a sample of more than 400 press titles, does not exclude the possibility that a bias may be fully explained by the editorial line wanted by the owner.

In 2018 (Garz et al. 2018), Marcel Garz, Gaurav Sood, Daniel Stone and Justin Wallace published a study and used the title of Gentzow and Shapiro’s 2010 article to expand on its conclusions. They base their analyses on the U.S. presidential campaigns of 2012 and 2016. Their results are in the same direction, that is, the bias practiced by the media corresponding to the political positions of readers, rather than the ideological choices of the owners. The bias is also reinforced by the development of social networks.

Francesco Drago, researcher at the University of Naples, and Tommaso Nannicini and Francesco Sobbrio, two researchers at Bocconi University in Milan, (Drago et al. 2014) are directly interested in the number of press titles and question the link between newspaper competition and the quality of political monitoring. They analyze the daily local press in Italian cities with more than 15,000 inhabitants¹² over the period 1993–2010. The researchers show that the number of press titles is positively correlated with the performance of municipalities measured by various indicators, such as tax collection. In other words, a competitive supply of newspapers also puts ideological biases into competition and improves the overall quality of information. They also analyze an interesting effect of electoral regulation: a mayor who is constrained by a time limit on mandates is less efficient in his last mandate. The study also shows that the local press is more effective on local elected officials than the national press.

In 2013, Andréa Prat and David Stromberg published The Political Economy of Mass Media (Prat and Stromberg 2013), the term political economy being used here in the sense of economic analysis of public decisions (Magis 2016)¹³. The authors develop a theoretical approach and summarize the previous scientific study. They insist on an informational bias that is not to be neglected, consisting of highlighting information of no political importance in order to divert the attention of voters from essential political decisions. They also show that ideological bias has more influence on unstable voters and when there is little competition between the media, in other words, when the bias promoted by one media outlet is not challenged by other media. They conclude their summary with three statements: 1) media scrutiny increases political accountability; 2) media pluralism and a healthy commercial motive are effective defenses against media capture; and 3) voter information and voting outcomes are affected by the media.

Their conclusion is that the press or media are particular companies, different from others in the sense that competition plays a different role. In addition to the traditional effect on price or quality, competition between media also determines the coverage of public decisions and thus partially explains both the effectiveness of public policies and the electorate’s vote.

The two researchers had conducted a similar study a few years earlier on the development of commercial television in Sweden. They tested voter behavior in two elections in 1988; one before the arrival of commercial television, and one after, in 1991 (Prat and Stromberg 2005)¹⁴. The result is an increase in the consumption of political information, mainly among those who did not use it during public television. More recently, Sebastian Elligsen and Oystein Hernaes (Elligsen and Hernaes 2015) measured the impact of commercial television on the effectiveness of public policies, but this time in Norway.

Cagdas Agirdas (Agirdas 2015), a professor at Tampa University in Florida, is studying the impact of the sharp decline in the number of newspapers in many U.S. cities. This recent decline is due to the decrease in resources, which is in turn due to the arrival of the Internet. This rapid change in the situation makes it possible to make a comparative analysis of the content of press articles when the number of newspapers was still large and since its decrease. For this purpose, the author studies the archives of 99 press titles over the period 1990–2009. He is particularly interested in the way in which information is reported on the evolution of the unemployment rate. As other authors have shown, there is an ideological bias in all newspapers. Thus, conservative (Democrat) newspapers publish 17.4% (12.8%) additional articles on unemployment when the President is a Democrat (Republican) and when the newspaper has a competitor in the same trading area. In the same area, with a sociology of voters that has not changed, the number of articles in the same context decreases considerably (respectively, 3.5% for articles published in conservative newspapers and 1.1% for Democratic newspapers). This means that the strategy of newspapers is oriented by the search for readership, rather than by an ideological political positioning. When there are several newspapers in the same market, they adopt a strategy of differentiation, and when there is only one newspaper left, the ideological bias decreases in order to have the widest possible readership and to attract former readers of the disappeared newspaper(s).

Fernanda De Leon (De Leon 2016) analyzed 90 U.S. newspapers during the election campaigns for the governor, attorney general and senator elections in 2002 and 2006, that is, 154 campaigns. The study covers seven states (California, Florida, Michigan, Ohio, Oregon, Texas and Wisconsin), including 658 counties. The results of the study are that ideological bias can occur when newspapers are in competition: the bias which is chosen then depends on the ideological position of the readers.