After All This Time? The Impact of Media and Authoritarian History on Political News Coverage in Twelve Western Countries

Abstract

Historical classifications of journalistic traditions are the backbone of comparative explanations for political news coverage. This study assesses the validity of the dominant media systems framework and proposes and tests a novel framework, which states that a history of authoritarianism affects today’s coverage. To facilitate a clean cross-national comparison, we focus on the same person and measurement in 12 Western democracies, that is, the use of the pejorative terms “sexist,” “racist,” “dictator,” and equivalents to describe Donald Trump. Our manually validated automated content analysis (2016–2018; N = 27,830) shows that content varies along with countries’ media and authoritarian history: pejoration is more common in countries with a polarized pluralist media system and former authoritarian countries than elsewhere. Newspapers’ ideology does not matter, irrespective of countries’ level of political parallelism or experiences with authoritarianism. Combined, we provide new methodological and theoretical handles to further comparative communication research in Western democracies.

Why does news appear in different forms in different countries? In Four Theories of the Press (1956), Fred Siebert, Theodore Peterson, and Wilbur Schramm proposed that what we read in the paper today is the product of a historical interplay between press, government, and society. This work would later inspire Daniel Hallin and Paolo Mancini’s (2004) seminal study Comparing Media Systems, which argued that the historical development of media systems shapes content features of coverage. Since then, Hallin and Mancini’s classification has been the most prominent and virtually uncontested comparative framework (e.g., Benson, 2004; Strömbäck & Luengo, 2008). However, discouraged by the lack of standardized measurements (Norris, 2009) and growing concerns over their relevance in a context of global convergence (Blumler & Gurevitch, 2001; Hallin & Mancini, 2012), various other important historical differences between countries have remained unexplored. In this study, we propose and demonstrate empirically that countries’ experience with authoritarianism is an equally viable explanation for what we read in the papers today. In particular, we contend that these experiences leave such deep-seated national traumas that they serve as recurring frames of interpretation in contemporary news coverage.

The purpose of this study is to assess the impact of countries’ media and authoritarian history. To this end, we develop a highly standardized design, which holds the topic and measurement constant across all countries under investigation. We do so by focusing on the use of pejorative terms that are known to provoke a sense of disgust in all established democracies, such as “sexist,” “racist,” “dictator,” and equivalents in news coverage of one single person. Following Hallin & Mancini’s (2004) work, we argue that in countries where journalistic standards promote a detached style of writing, journalists are more likely to avoid these terms than elsewhere. Furthermore, we develop a novel theory, which is based on insights from political science literature on authoritarian legacies (e.g., Art, 2005; Costa Pinto, 2010; Dinas & Northmore-Ball, 2019). Building on these insights, we argue that journalists are more likely to produce pejorative coverage when the object of coverage is associated with historical examples of authoritarianism.

Given its comparative angle, this study speaks to several longstanding debates in communication science. Theoretically, our study adds a novel historical classification to an area with a relatively limited number of theoretical contributions (Norris, 2009). More importantly, by doing so, we demonstrate the theoretical fertility of explanations focusing on countries’ political culture (see Gurevitch & Blumler, 2004; Hallin & Mancini, 2004). Empirically, this study provides a more comprehensive validation of Hallin and Mancini’s classification than earlier efforts. We do so by expanding the geographical scope from a limited number of prototypical examples of each media system (six at most, see, e.g., Esser & Umbricht, 2013; Strömbäck & Luengo, 2008) to 12 countries. This expansion enables us to assess the viability of this classification beyond prototypical cases.

Methodologically, we address three problems typically associated with comparative analysis of media coverage. That is, our focus on coverage of a single person who (a) has attracted extensive media attention in many countries, (b) has been frequently labeled with pejorative terms, and (c) has been recurrently compared to historical and contemporary authoritarian figureheads, enables us to achieve higher levels of sample, measurement, and instrument equivalence than prior research. Currently, US President Donald Trump is the only case that satisfies these criteria. We use the frequent pejoration of Trump to our advantage to conduct a systematic, manually validated automated content analysis of 27,830 articles in 35 newspapers in 12 Western democracies (2016–2018).

Theory and Hypotheses

Journalistic neutrality and news content

In the production of news content, media practitioners must choose between two opposing roles (Cohen, 1963). They may assume an active role, aimed at influencing the public through providing interpretation, commentary, and criticism. Alternatively, they may consider a neutral role, aimed at informing the public by providing impartial and objective coverage. Although both roles have their merits, the weight of evidence is that media practitioners everywhere pledge allegiance to the neutral role (Hallin & Mancini, 2004; Tuchman, 1978; Weaver, 1998). However, they may still move toward a more active role when they feel pressured to do so. To what degree this is the case may be the result of various internal and external pressures resulting from journalists’ personal beliefs, media routines, organizational characteristics, extra-media influences, or societal influences (Shoemaker & Reese, 1991).

In the following paragraphs, we argue that countries’ media and authoritarian history establishes pressures to adhere to or abandon this coveted standard of neutrality. Although the standard of neutrality influences various features of news coverage—such as the use of frames, reporting styles, and the promotion of political agendas—we focus on the use of the pejorative terms “sexist,” “racist,” “dictator,” and equivalents. Even when justified, such words are deeply discrediting and imply that an actor’s behavior is beyond the pale. The use of these terms, therefore, arguably requires a conscious decision on the part of the journalist.

Explaining cross-national differences in coverage

Media history

The prevailing comparative explanation of media coverage is discussed in Hallin and Mancini’s (2004) book Comparing Media Systems. Among other things, these scholars argue that news content should be viewed as the outcome of countries’ media history. One such outcome is the promotion of a detached style of writing. The prevalence of this style of writing can be traced back to two historical characteristics: the professionalization of the press and the amount of state intervention in its development.

In Anglo-Saxon countries, the press was primarily left to market forces and retained independence from the state (Hallin & Mancini, 2004). Journalism became a profession with its own educational, organizational, and normative structure, all emphasizing the distinction between news and opinion (Schiller, 1981). This development resulted in a “liberal” media system and a detached, information-oriented style of journalism. Although press in continental Western Europe also experienced a process of professionalization, the press did retain strong ties to the state and politics. These developments resulted in “democratic corporatist” media systems, in which the emphasis on neutrality was weakened in relation to political advocacy and dedication to the public interest. The development of the press in Southern Europe, by contrast, followed a different historical trajectory. Here, the strong dependence on parties and the contributions of skilled writers and politicians resulted in the development of “polarized pluralist” media systems.

It is plausible that countries’ media system affects the use of pejoration in news content. In liberal systems, media experience a strong societal pressure to act as a neutral observer. The weak ties to the state and politics furthermore limit the influence of external actors. Their high levels of professionalism may also establish intramedia pressures to provide neutral coverage, for example, resulting from a code of ethics, or journalists’ self-conception as detached observers (Kepplinger & Köcher, 1990). Pejoration is, therefore, likely to be avoided or to be edited out afterward. In democratic corporatist systems, the neutral role coexists with an active role. Governments, the public, and journalists themselves may expect the media to abandon the standard of neutrality to defend the public interest. The motivation to use pejoration is, therefore, mixed. In polarized pluralist systems, the pressure to assume an active role arguably outweighs the pressure to remain neutral. Society, political parties, and journalists themselves may agree that the media must provide commentary and abandon the standard of neutrality when it is deemed appropriate. The production and publication of pejorative content are thus less objectionable than elsewhere.

Various studies show that role conceptions of journalists as detached observers are most common in countries with a liberal media system and least common in countries with a polarized pluralist system (Donsbach & Patterson, 2004; Köcher, 1986; Van Dalen, Albæk & De Vreese, 2011). Studies drawing on comparisons of news coverage furthermore demonstrate that countries’ media system affects the prevailing style of journalism. These studies show that opinionated reporting styles (Esser & Umbricht, 2013), critical news content (Benson, 2010; Benson & Hallin, 2007), and interpretative styles (Strömbäck & Dimitrova, 2006) are most common in prototypical examples of polarized pluralist systems and least common in examples of liberal systems. Tied back to pejoration, we expect that:

H1a: Pejoration varies across media systems and is most common in polarized pluralist systems and least common in liberal systems.

Countries’ media history may also influence the role of newspapers’ ideology. The concept “party-press parallelism” was first coined by Seymour-Ure (1974) to describe the close alignment of parties and press in Britain. Hallin and Mancini (2004) later use the concept “political parallelism” to describe the general bonds between press and ideologies. Parallelism is strongest when newspapers defend only one political–ideological current and weakest when they remain fully impartial. Strong parallelism is typically found in Southern Europe, medium levels in continental Western Europe, and weak parallelism in most Anglo-Saxon countries.

Parallelism influences news coverage, because it determines how newspapers respond to ideas of other ideological currents. When parallelism is weak, newspapers aim to provide balanced access to different ideological voices. When parallelism is strong, newspapers only present their ideology. Ideological diversity can only be achieved insofar different papers offer different views (Hallin & Mancini, 2004). Parallelism may also produce additional pressures to use or refrain from using pejoration. For instance, parallelism implies a firm hold of political parties on the media, which encourages the production of partisan content. It also translates to internal pressures to produce partisan content, because poorly paid jobs in journalism serve as a springboard to a career in politics (Ortiz, 1995), and because it promotes journalists’ self-conceptions as political advocates (Hallin, 1986). It is therefore plausible that newspapers’ ideology matters more in countries with high levels of parallelism than elsewhere.

Although scholarship agrees that parallelism affects role conceptions of journalists as political advocates (e.g., Donsbach & Patterson, 2004; Van Dalen et al., 2012), evidence that this spills over to news content is mixed (Benson & Hallin, 2007; Tandoc et al., 2013; Van Dalen et al., 2012). Despite this mixed evidence, it is likely that the higher the level of parallelism, the more newspapers’ ideology matters. If this is true, the difference in the prevalence of pejoration between, for instance, the Canadian Toronto Star (left) and the National Post (right)—published in a country with low levels of parallelism—is less pronounced than that between the French papers Le Monde (left) and Le Figaro (right). We, therefore, expect that:

H1b: The higher countries’ level of political parallelism, the more pronounced the difference in pejoration between left- and right-leaning newspapers.

Authoritarian history

Countries’ authoritarian history may also contribute to what degree it is deemed acceptable to use pejoration. Within the context of Western democracies, historical experiences with right-authoritarianism are arguably especially important determinants of media content. That is, in the interwar period, fascism was the leading ideology in Austria, Germany, and Italy. Later in the century, right-authoritarianism retained its significance in the form of military dictatorships in Spain, Portugal, and Greece. These regimes were notorious for the intensity of their well-publicized physical repression, surveillance, and propaganda (Dinas & Northmore-Ball, 2019), thereby leaving deep-seated collective traumas.

It is commonly acknowledged that historical experiences with right-authoritarianism have resulted in strong pressures to create a rupture with the past. Institutional pressures include constitutional provisions permitting the criminalization and prosecution of parties and leaders of the past regime (Bourne, 2018; Costa Pinto, 2010; Morlino, 2010). Societal pressures are even further reaching and extend to anyone who might be associated with the past regime. Even the slightest similarity with the authoritarian predecessor may be used as an excuse to recall the traumatizing records of the past. As a result, the past is frequently recalled in elite and public debate to discredit opinions, persons, and parties (Encarnación, 2004; Morlino, 2010).

It is plausible that the authoritarian past also creates pressures to produce pejorative news content. Good examples of institutional pressures are the Italian and Portuguese constitutional charters, which were designed to counter all remnants of the past regime (Costa Pinto, 2010; Morlino, 2010). Evidence for societal pressures can be found in Spain, where for years, the mainstream right was deeply mistrusted due to its perceived association with the Franco regime (Encarnación, 2004; Morlino, 2010). The attempts of the Portuguese center–right parties CDS Partido Popular (CDS-PP) to accuse the left of (Santana-Pereira et al., 2016) authoritarian politics furthermore shows that everyone can be targeted (Santana-Pereira, Raimundo, & Costa Pinto, 2016). The need to produce content that resonates well with the public (Snow & Benford, 1988) and journalists’ self-conceptions as defenders of democracy may furthermore constitute important internal pressures to target potential threats to democracy.

Empirically, this assertion can be loosely substantiated by arguments made in other studies in the field of communication science. First, some studies have argued that past experiences with authoritarianism have resulted in an emphasis on the promotion and defense of democratic values in news media (Gunther, Montero, & Wert, 2000; Hallin & Mancini, 2004; Van Dalen et al., 2012). In keeping with this argument, Köcher (1986) shows that journalists in former authoritarian Germany are almost twice as likely to agree that it is their task to oppose antidemocratic parties as their British counterparts. Thus, we expect that:

H2a: Pejoration is more common in former authoritarian countries than elsewhere.

Authoritarian legacies may also mitigate the impact of newspapers’ ideology, irrespective of countries’ level of parallelism. From literature on party politics, we know that the institutional and societal pressures discussed before are especially strong for parties with a higher risk of being associated with the past (Art, 2005; Van Spanje, 2018). A similar argument can be made for outlets with a more similar ideological leaning as the authoritarian predecessor. These outlets may be aware of the negative impact nonpejorative content of controversial figures may have on their public image. Even on a subconscious level, journalists may feel inclined to discredit controversial figures, because they grew up in a context where it was common to do so as well. In former right-authoritarian countries, left-leaning newspapers, therefore, have an ideological motivation to discredit anyone associated with the past regime, whereas right-leaning papers have a pragmatic reason.

Literature on party politics and transitional justice corroborates that discrediting the authoritarian past transcends the division between left and right. The German center-right party Christlich Demokratische Union Deutschlands (CDU), for example, has made considerable efforts to disassociate itself from the Nazi past (Art, 2005; Van Spanje, 2018). Likewise, the Portuguese CDS-PP still excludes anyone who is associated with the Estado Novo regime (Costa Pinto, 2010). If the same applies to newspapers, it is plausible that ideology matters less in former right-authoritarian countries, because both right- and left-leaning outlets have an interest in discrediting anyone associated with the past regime. The empirical implication is that the difference in the prevalence of pejorative coverage between DieTageszeitung (left) and Die Welt (right) in former right-authoritarian Germany is less pronounced than that between de Volkskrant (left) and Algemeen Dagblad (right) in the Netherlands. In short, we expect that:

H2b: The difference in the prevalence of pejorative coverage between left- and right-leaning newspapers is less pronounced in former authoritarian countries than elsewhere.

Methods¹

Case selection: Donald Trump

Our case selection is based on theoretical and methodological criteria. Theoretically, the aim of this study is admittedly challenging. Not only do we wish to validate a framework that has passed various empirical tests already, but we also propose a novel framework that is yet to pass its first test. For both purposes, a least-likely case selection is desirable. In a least-likely case, all dimensions of a case predict that an outcome will not occur, but “if the theory turns out to be correct regardless, the theory will have passed a difficult test, and we will have reason to support it with greater confidence” (King, Keohane, & Verba, 1994, p. 209). As many have pointed out, a least-likely case study is especially valuable at the stage that candidate theories, like ours, are being tested. It also permits us to evaluate whether established theories, like Hallin and Mancini’s, are capable of passing a more rigorous test. As the only foreign politician receiving extensive media attention in all countries under investigation, US President Donald Trump presents a unique least-likely case to study cross-national differences in coverage. That is, diffusion theory predicts that news media in different countries should behave in a very similar manner covering foreign news. The reason for this is that media often draw information from the same sources for such news, such as international news agencies. This is in stark contrast with coverage of domestic politicians, which constitutes the majority of political news and for which newspapers collect their own information. It is therefore plausible that if we find evidence for cross-national differences in coverage of Trump – and our explanations for these differences – cross-national differences in political news coverage at large are much more pronounced.

Aside from its theoretical advantages, a focus on Trump also helps us resolve three problems resulting from the limited comparability of media data across countries. In survey research, these problems are qualified as sample, measurement, and instrument inequivalence. First, focusing on a single person enables us to hold the characteristics of the object of coverage constant across countries, thereby improving the sample equivalence of our data. Second, the repeated pejoration of Trump as “sexist,” “racist,” “dictator,” and equivalents permits us to employ a measurement that is understood in the same way in all countries under investigation. This allows us to achieve higher levels of measurement equivalence. Finally, a focus on Trump offers an opportunity to achieve higher levels of instrument equivalence by paying special attention to comparisons with notoriously despised authoritarian figureheads. By doing so, we minimize the bias introduced by the use of different coders and dictionaries across countries.

Data

Several criteria guided the data collection. First, the selection was constrained by the online availability of news sources in the databases Nexis Uni and Go Press Academic. We selected countries in such a way to facilitate considerable variation in countries’ media and political history. Within these countries, we selected all available national quality newspapers, as to ensure that we compare the same type of newspapers in all countries.² We then retrieved all available coverage mentioning Trump between 1 January 2016, and 31 December 2018. Finally, we ensured that we study a time frame in which news coverage was available in all countries by narrowing down our selection to articles published after the date of the announcement of Trump’s candidacy on 16 June 2016. This procedure resulted in 27,830 articles in 35 newspapers in 12 countries.

Dependent variable: Pejoration

The dependent variable of this study is the pejoration of Trump. To detect the use of pejoration, we first conducted a systematic automated content analysis, based on an extensive dictionary of pejorative terms. This dictionary was translated by native speakers to seven languages, covering the 12 countries under investigation. Words were considered pejorative if they implied a comparison or association with political currents generally considered beyond the pale. This broadly includes (a) antidemocratic currents (e.g., “authoritarian” and “dictator”), (b) illiberal beliefs that deny the equality between citizens (e.g., “sexist” and “racist”), (c) historical examples of authoritarian regimes (e.g., “fascism” and “Benito Mussolini”), and (d) contemporary examples (e.g., “Neo-Nazism” and “Vladimir Putin”).

The automated content analysis returned 16,991 hits spread across the 27,830 articles in our dataset. To redress the chances of articles being incorrectly coded as positive, we asked our coders to validate each hit. We did so by presenting them with short text fragments (snippets) in which the captured term and Trump’s name were capitalized. Our coders were asked to evaluate whether the capitalized term was indeed pejorative, as to identify incorrectly captured words. In Italian articles, for example, the search string “Nazi” incorrectly returned the word “nazionale” (national). We then asked whether the term was linked to Trump through a label, a comparison, or a general association. In this phase, texts such as “Trump meets with authoritarian leader Kim Jong Un” were recoded as negative. Finally, we asked all coders to code the same subset of English snippets (N = 320), which confirmed that coders worked according to the same criteria (Krippendorff’s alpha = 0.75).³ Ultimately, these endeavors resulted in a dependent variable where “1” indicated that an article contained pejorative language in relation to Donald Trump and “0” that it did not.

Independent variables

Drawing on the classification proposed by Hallin and Mancini (2004), we distinguish between countries with (a) liberal, (b) democratic corporatist, and (c) polarized pluralist media systems. Building on the same work, we furthermore differentiate between countries with (a) low, (b) medium, and (c) high levels of political parallelism. To investigate the influence of countries’ authoritarian history, we classify countries according to their prior experiences with right-authoritarianism. On the level of the news outlet, we identified newspapers as (a) left-leaning, (b) centrist, or (c) right-leaning. Finally, we control for the length of the article, because pejorative coverage is more likely to occur in longer articles and because the average length of an article may vary between countries. All country and newspaper characteristics are summarized in Table 1.

Analysis Strategy: Bayesian Multilevel Logistic Regression

Since countries are the main unit of analysis, the prime statistical challenge is producing an adequate estimation of country-level effects. In the empirical part of this manuscript, we have made two methodological choices to address this challenge. First, we employ multilevel analysis techniques, with articles (Level 1) nested in outlets (Level 2) and countries (Level 3). These techniques take into account the variance explained by the clustering of observations within outlets and countries. In addition, multilevel techniques are commendable for their ability to estimate interactions between different levels of clustering or “cross-level interactions.” They do so by allowing the slope of an effect at a lower level of clustering (in our case newspapers’ ideology) to vary across countries. Given that centrist newspapers were not available in all groups under investigation, we did not include centrist newspapers in analyses estimating these interactions.

Second, we address difficulties arising from the fact that we are dealing with a small number of countries (N = 12) spread across two or three groups. Using frequentist multilevel approaches would be problematic because when the number of countries is small, the estimation of variance components, point estimates and confidence intervals tend to be biased with up to as much as 20% (Stegmueller, 2013). Overall, these techniques would substantially increase the chances of making a Type I error. In such cases, various studies have recommended the use of Bayesian analysis techniques (Baldwin & Fellingham, 2013; Stegmueller, 2013), which have shown to produce unbiased estimates with as little as three clusters. They do so by estimating a series of parameters and creating a density distribution (or “posterior distribution”) of all credible parameter values.⁴ In addition to avoiding crude measures such as significance tests (see Levine et al., 2008), Bayesian hypothesis testing allows for an intuitive interpretation and is merely a way of expressing the credibility of a hypothesis, in view of the data. The credibility is calculated as the share of the posterior distribution that supports the hypothesis. For instance, if a hypothesis predicts that a particular effect is negative, the empirical support for this hypothesis equals the percentage of the density distribution that falls below the value zero on the x-axis. To allow for a substantive reading of the results, we report a credible interval (CI) containing the 90% parameter estimates that are best supported by the data.

Results

Mapping Cross-National Differences

Figure 1shows the amount of pejorative coverage as a percentage of the total coverage of Trump in each country, with darker colors indicating a higher percentage. This figure shows that pejorative coverage is common, ranging between 18.26% of total coverage in the Netherlands and 47.01% in Spain. Cross-national differences seem to reflect a clear geographic divide, with pejoration being more common in Southern Europe than elsewhere. However, this distinction does not capture all variation. For instance, despite the geographic proximity of the Netherlands and Germany, only 18.26% of Dutch news coverage contains pejoration, whereas in Germany, this equals 34.12%.

Explaining Cross-National Differences

Media history

The first explanation for these cross-national differences held that the use of pejoration varied across media systems (Hypothesis 1a). The main effect of countries’ media system (Table 2, Model 1a) evaluates whether this is the case. The negative value of the dummy for democratic corporatism (β = −0.165) tells us that pejoration is less common in countries with this system than in countries with a polarized pluralist system. This coefficient indicates that this difference equals 4.07 percentage points.⁵ Likewise, the dummy for liberal systems (β = −0.176) suggests that pejoration is 4.39 percentage points less common in countries with this system than in countries with a polarized pluralist system.

To assess the credibility of Hypothesis 1a, we conduct three tests on the posterior distributions of the coefficients for countries’ media system (Figure 2a). To test whether pejoration is more common in polarized pluralist systems than elsewhere, we calculate the share of each of the two distributions falling below zero. This reveals 99% support for the expectation that pejoration is more common in polarized pluralist systems than in democratic corporatist systems and 99% support for the expectation that it is more common than in liberal systems. A final test calculates the credibility of the expectation that pejoration is more common in democratic corporatist systems than in liberal systems. This test provides only 60% empirical support for this expectation. Thus, we find considerable (but not full) support for Hypothesis 1a.

A second expectation was that higher levels of political parallelism would result in more pronounced differences between left- and right-leaning newspapers (Hypothesis 1b). We test this by estimating an interaction between countries’ level of parallelism and newspapers’ ideology (Model 1b, Table 2). The negative value of the main effect of ideology (β = −0.026) predicts that pejoration is 0.64 percentage points less common in right- than in left-leaning newspapers in countries with low levels of parallelism. However, the low value of the interaction term for medium levels of parallelism (β = 0.041) and the near-zero value of that for interaction term for high levels (β = 0.023) show that there is little reason to believe that this ideological difference is more pronounced in countries with higher levels of parallelism.⁶

Hypothesis tests based on the posterior distributions of the interaction terms (Figure 2b) confirm this preliminary conclusion. These tests reveal 70% support for the expectation that newspapers’ ideology matters more in countries with high levels of parallelism than in countries with low levels and 62% support for the expectation that it matters more than in countries with medium levels. In addition, there is very little support (36%) for the expectation that these ideological differences are more pronounced in countries with high than in countries with medium levels of parallelism. The data, therefore, provide little to no support for Hypothesis 1b.

Altogether, these findings suggest that differences in news content can, to some extent, be attributed to countries’ media history. The determination coefficient for Model 1a reveals that this model explains 8.99% of the variance in our data, of which approximately 3.70% is accounted for by countries’ media system. At the same time, the near-zero value of the partial determination coefficient for the interaction term between newspapers’ ideology and countries’ level of political parallelism (Model 1b) shows that countries’ media history cannot account for differences in content between newspapers of different ideological leanings.

Authoritarian history

We furthermore argued that pejorative coverage would be more prevalent in former authoritarian countries (Hypothesis 2a). We test this by estimating a model with a dummy variable for countries’ authoritarian history (Model 2a, Table 2). In keeping with Hypothesis 2a, the positive coefficient for countries’ authoritarian legacy (β = 0.106) shows that pejoration is around 2.62% more common in former authoritarian countries. A test based on the posterior distribution of this coefficient (Figure 2c) reveals that there is considerable support for Hypothesis 2a (95%).

Our final hypothesis was that past experiences with authoritarianism would mitigate the impact of newspapers’ ideological leaning (Hypothesis 2b). We test this by estimating an interaction between newspapers’ ideology and countries’ authoritarian history (Table 2, Model 2b). The near-zero value of the main effect of newspapers’ ideology (β = 0.007) shows that in countries without a legacy of authoritarianism, pejoration is almost equally common in right- and left-leaning newspapers (the difference is less than 1 percentage point). Countering our hypothesis, the near-zero value of the interaction term between newspapers’ ideology and countries’ authoritarian legacy (β = 0.020) suggests that it is unlikely that this difference is more pronounced in former authoritarian countries. A hypothesis test based on the posterior distribution of this analysis (Figure 2d) confirms that there is indeed little empirical support for Hypothesis 2b (68%).

These analyses show that this novel way of classifying countries is able to pass a difficult test. What is more is that its predictive capacity is comparable to that of Hallin and Mancini’s classification. That is, the partial determination coefficient for countries’ authoritarian legacy presented in Model 2a, shows that 3.70% of the variance in the data is accounted for by countries’ authoritarian legacy. This is only 0.01% less than the partial determination coefficient for countries’ media history presented in Model 1a. In other words, this novel classification performs equally well as Hallin and Mancini’s classification of media systems despite being considerably more parsimonious. At the same time, this explanation is equally incapable of accounting for differences between newspapers of different ideologies.

Robustness test

Our design already enables a high level of cross-national comparability. However, the use of different coders and dictionaries in different countries may be a source of instrument inequivalence. The reason for this is because there may be cultural, linguistic, and semantic differences across languages and coders. For instance, the number and type of adjectives used may very well be culturally determined. Likewise, some languages have a much richer vocabulary than others, resulting in variation in terms of the length of our dictionaries. Finally, in terms of semantics, it can be debated whether words such as “authoritarian” or “bigot” are equally pejorative in all languages.

To ensure that our findings for countries’ media system and authoritarian history cannot be attributed to this possible lack of instrument equivalence, we conduct a test that focuses on two subtypes of pejoration, namely comparisons with historical and contemporary examples of authoritarianism. These types of pejoration are less sensitive to cultural, linguistic, and semantic influences because (a) there is virtually no cross-national variation in the number of synonyms for names such as “Hitler” or “Putin” and words such as “Nazism” and “fascism” and (b) they leave substantially less room for interpretation than other forms of pejorative coverage. Figure 3 visualizes the results of this robustness test.

The left panel of Figure 3a shows that our findings for historical pejoration mirror the patterns of earlier findings, with pejoration being more common in polarized pluralist systems than elsewhere. Likewise, we find little to no difference between liberal and democratic corporatist systems. By contrast, our findings do not hold when focusing on contemporary pejoration, which is equally common in polarized pluralist systems as liberal systems. Figure 3b shows that our findings for countries’ authoritarian history do hold. Pejoration is systematically more common in countries with a legacy of authoritarianism than elsewhere, regardless of whether it concerns contemporary or historical pejoration. In short, Hypothesis 2a is robust to this particular test, whereas this is less so for Hypothesis 1a.

Discussion

In Four Theories of the Press, Siebert et al. (1956) first asked why news content appears in different forms in different countries. A few decades later, the landmark study of Hallin and Mancini (2004) would attempt to formulate an answer to this question, arguing that cross-national differences should be attributed to the historical development of countries’ media system. Validation of this framework, as well as the development of new ones, however, has remained a difficult theoretical and empirical task. In this study, we developed a least-likely and standardized test to investigate whether countries’ media and authoritarian history affect content features of news coverage. This enabled us to demonstrate empirically that aside from countries’ media history, historical experiences with authoritarianism influence what we read in the paper today. We found that pejorative coverage is more common in countries with polarized pluralist media systems and former right-authoritarian countries than elsewhere. At the same time, we found little evidence that newspapers’ ideological leaning matters: pejoration appeared to be equally common in left- and right-leaning newspapers, regardless of countries’ level of political parallelism or past experiences with right-authoritarianism.

The findings of this study play well to at least three longstanding debates in communication science. Theoretically, we advanced a novel explanation for cross-national differences in news coverage. In particular, we argued that countries’ traumatic historical experiences with right-authoritarianism would influence political news content. Our theoretical contribution is also relevant to the field of comparative communication at large. The form in which news content appears is different in every country and outlet. This characteristic makes studying macro-level determinants of news content especially instructive. They sensitize us to the role systemic characteristics play in the production of news content in a way that single-country research cannot. This is where the broader contribution of this study lies: in revealing the theoretical fertility of what comparative scholars have identified as the main area of theoretical expansion, that is countries’ political culture (see Gurevitch & Blumler, 2004).

Empirically, our study shows that the two historical classifications we discuss are able to pass an extremely difficult test. In light of this evidence, we can conclude that countries’ history still accounts for cross-national variation in news coverage. This counters two recurring criticisms fielded against Hallin and Mancini (2004), namely, their inability to validate their conceptualizations empirically (see, e.g., Esser & Umbricht, 2013; Norris, 2009) and their inappropriateness in times of global convergence (Blumler & Gurevitch, 2001; Hallin & Mancini, 2004, 2012). Indeed, the finding that pejoration is equally common in liberal as in democratic corporatist systems is consistent with the argument that some media landscapes are converging toward the “liberal model” (Hallin & Mancini, 2004, 2012). However, the sharp contrast we observed between these two groups of countries on the one hand and polarized pluralist systems, on the other hand, provide validation for (parts of) Hallin and Mancini’s (2004) classification. We also showed that countries’ authoritarian history, a framework based on insights from political science, may provide a more robust explanation than this landmark classification. Even more so, this classification performs equally well in terms of explanatory power, despite being considerably more parsimonious.

Methodologically, this study addressed several recurring challenges resulting from a limited comparability of media data (Gurevitch & Blumler, 2004; Norris, 2009). Prior research has already taken significant steps forward by focusing on news coverage of comparable objects to increase the comparability of the data (see Kaid & Holtz-Bacha, 1994), and using random sampling techniques to improve its representativeness (most notably Esser & Umbricht, 2013). Our study shows that a careful case selection may even further improve the comparability of analysis of news content. First, we improved the sample equivalence by focusing on a narrow topic to hold the object of coverage constant across countries. Second, our focus on a limited set of words with similar connotations in all countries enabled us to reach higher levels of measurement equivalence. Finally, acknowledging a possible bias introduced by a lack of instrument equivalence, we conducted robustness tests focusing on forms of pejoration that are virtually insensitive to cultural, linguistic, and semantic differences. Our efforts to take these considerations into account meet the growing demand for a methodological toolkit to permit systematic comparative analyses in communication science (Norris, 2009; Wirth & Kolb, 2004). Even more so, the dataset and content analyses compiled for this specific study can very well be utilized by future scholarship interested in macro-level effects on news coverage.

Notwithstanding these contributions, several limitations and avenues for future research have to be considered. Perhaps the most pressing theoretical limitation is the generalizability of our theoretical arguments beyond the context of Western democracies. Like Hallin and Mancini’s work, our novel framework is grounded in several implicit assumptions, including that the press is free and that historical experiences with authoritarianism have been sufficiently impactful to leave a collective trauma. Only if these assumptions are met in countries other than those included in this study, we can speak of a truly generalizable framework. If this is not the case, then the usefulness of our frameworks is limited to its classificatory function. This necessarily brings us to a second, empirical limitation. Although large for comparative communication standards, the number of countries we study does not suffice to add nuance to our empirical models. This is, for instance, reflected in our choice to classify countries into generic classes, such as “right-authoritarianism” or “polarized pluralist systems.” In addition, this limited scope has made that we were unable to study the interaction between countries’ media and authoritarian history. Yet, several studies (e.g., Gunther et al., 2000; Hallin & Mancini, 2004) underline that media system formation and countries’ authoritarian history is intertwined in some countries, and not so much in others. Further expanding our database to more countries, more topics, and more diverse measurements may resolve these limitations.

Finally, our methodology also suffers from several limitations. First, it is important to acknowledge a possible constraint on the replicability of our findings. At this point, no politician other than Trump has received this much negative attention in this many countries. That said, it is conceivable that a similar opportunity will arise in the future, as several countries have elected a leader whose democratic credentials are widely doubted (e.g., Boris Johnson, Jair Bolsonaro, Viktor Orbán) and about whom negative coverage is currently accumulating. A second methodological limitation is a direct consequence of our standardized design. By removing various sources of variation both between newspapers and between countries, the effects reported in this study are likely to be underestimated. This may very well explain our null findings for our hypotheses on ideological differences between newspapers. Third, our focus on a single case hinders us in our ability to say something about how much countries’ history matters. News coverage on Trump may not be representative of other news coverage. This is especially consequential for our novel theoretical framework, for which this study presents the first and only test. For this framework, our contribution mainly lies in demonstrating its theoretical and empirical viability, although the results presented in this study should be considered exploratory and preliminary. More research is necessary to assess its validity beyond this case. A final limitation arises from our decision to focus on a least-likely case. Our null findings for our tests of ideological differences suggest that a key assumption of a least-likely test is unfulfilled. That is, even a least-likely design rests on the assumption that the predicted outcome is possible (although improbable). Although it is theoretically possible to find ideological differences in coverage of Trump, there are two reasons to suspect that our design has crossed the fine line between improbability and impossibility. One reason is grounded in the fact that he is a foreign and notoriously unpopular politician. This may have rendered the role of newspapers’ ideology completely irrelevant. Another reason is that our focus on a single case makes that we lack an adequate benchmark to observe an effect if there is one. For instance, right-leaning newspapers may be more likely to produce pejorative coverage in general but refrain from doing so when it concerns coverage of a right-wing politician. Such alternative explanations cannot be ruled out unless we add another case to the analysis.

In spite of these shortcomings, our study provides reassurance that historical comparative classifications perform well in explaining news coverage. We demonstrated that after all this time countries’ history matters, always. Not only does this finding counter the most prominent criticisms of comparative scholarship exploring legacy effects on media content, but it also serves as an encouragement to expand the scope of theoretical work in this field. In this respect, the theoretical and methodological novelties presented in this study may provide a useful handle to guide these future endeavors.

Supporting information

Additional Supporting Information may be found in the online version of this article.

Please note: Oxford University Press is not responsible for the content or functionality of any supplementary materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

Footnotes

Acknowledgments

The authors extend their sincerest gratitude to Bert Bakker, Fernando Casal Bértoa, Cees van der Eijk, Tom Louwerse, Noam Lupu, Doug Rivers, Susan Vermeer, Stefaan Walgrave, Till Weber, Magdalena Wojcieszak, the four anonymous reviewers and editors Rousiley Maya and Lance Holbert for their feedback on earlier versions of this manuscript; Cecilia Badano, Gaspar de Bellefroid, Mark Eriksen, Álvaro González de Arrieta Martínez, Lara Heinz, Cecilia Badano, Gaspar de Bellefroid, Iris Schilder and, Katie Snyder, Lara Heinz, and Mark Eriksen for their impeccable work on the validation of the data; Damian Trilling for his help resolving computational problems in the data collection and processing phase of this project; Wouter de Nooy for his invaluable methodological advice; and Laura Jacobs, Lisanne Wichgers, Hans Beentjes, Sofie Marien, Ellen Claes, Marc Hooghe, Silvia Erzeel, Bart Kerremans, Stefaan Fiers, Heleen Touquet, Gunther Vanden Eynde and (one) other who must not be named (ML) for their involvement in or support for this or other works.

This work is generously supported by the Netherlands Organisation for Scientific Research (Grant 452-14-002 to Joost van Spanje), the University of Amsterdam School of Communication Research (ASCoR), and the University of Amsterdam Center for European Studies (ACES).

References