Winners, Losers, and the Press: The Relationship Between Political Parallelism and the Legitimacy Gap

Abstract

Recent work has explored how individual and institutional factors affect the gap in perceptions of political legitimacy between electoral winners and electoral losers, but has ignored the role that the political information environment, in general, and ideologically biased media, in particular, plays in exacerbating or diminishing this gap. By combining individual-level public opinion data in 28 countries, an expert survey on media systems, and a variety of country-level indicators, I find that higher levels of political parallelism in a country are associated with a larger winner-loser gap in institutional trust and satisfaction with democracy. The relationship is contingent on whether or not people are actually exposed to said media. This research, which links the study of political communication with the study of comparative political behavior, indicates that the increasing availability of partisan news around the world is a cause for concern.

Keywords:

• trust
• partisan news
• media bias
• legitimacy gap

Positive attitudes toward the political system are one indicator of a healthy democracy. Trust in political institutions and processes, in particular, bolsters a government’s ability to implement policies and increases the willingness of citizens to comply with those policies (e.g., Hetherington, 2005). Past research has demonstrated that trust in government is strongly related to the results of the last election (e.g., Anderson, Biais, Bowler, Donovan, & Listhaug, 2007; Esaiasson, 2011; Moehler, 2009; Moehler & Lindberg, 2009; Nadeau & Biais, 1993); those who voted for the winner consistently perceive institutions as more legitimate and trustworthy than do those who voted for the loser. The present research investigates how one aspect of the political information environment—the availability of ideologically biased media—mitigates or exacerbates these differences.

While biased media are believed to be related to attitudes toward policies and the political opposition, and political attitudes and out-party animosity are thought to be related to the legitimacy gap, no research, to date, has linked these two phenomenon. A few studies have looked at differences in perceptions of trust in the opposition and the media (e.g., Levendusky, 2013a; Mutz & Reeves, 2005). However, trust in the opposition is not as fundamental to democratic stability as trust in the system. The current research expands on past work, and using a variety of cross-national data, tests whether biased media is associated with a wider winner-loser gap.

After discussing the legitimacy gap and its theoretical relationship to political parallelism, I present the empirical evidence. I combine several cross-national data sets to examine perceptions of legitimacy between winners and losers in European countries that vary to the degree to which biased media are present. I find the gap in institutional trust as well as satisfaction with democracy is larger when ideological media are more ubiquitous. However, using coarsened exact matching (Iacus, King, & Porro, 2011), I show that that relationship is stronger among those who pay more attention to the news.

In total, this research highlights a third variable that impacts the legitimacy gap, which is measured neither at the individual level nor the traditionally defined institutional level, the two foci of legitimacy gap research. It demonstrates the importance of the political information environment, which filters how individuals respond to government and the degree of legitimacy they confer upon it. In doing so, this work adds to a small but growing literature on comparative political communication effects.

The Legitimacy Gap

Elections tend to increase trust in institutions and democratic processes across the electorate as a whole (e.g., Banducci & Karp, 2003; Price & Romantan, 2008). However, the effect is asymmetric: Citizens who voted for the winner tend to be more trusting than those who voted for the loser (Anderson & Guillory, 1997; Anderson et al., 2007; Keele, 2008). This gap in perceived legitimacy can be problematic for a number of reasons.

First, low levels of political trust among the losers leads to electoral instability. In the extreme case, losers pick up arms and civil war erupts. In the more ordinary case, bitter losers refuse to play the game and drop out of the electorate, turn to unconventional forms of political behavior, or are unwilling to compromise with the winners when necessary (Anderson et al., 2007; Moehler, 2009). Some grumbling, especially aimed at the other party, is expected after an election. However, at the end of the day, all citizens must view the electoral winners as the legitimate government, as “the viability of electoral democracy depends on its ability to secure the support of a substantial proportion of individuals who are displeased with the outcome of an election” (Nadeau & Biais, 1993, p. 553).

While low levels of perceived legitimacy among electoral losers are problematic, so are overly positive levels among electoral winners (Moehler, 2009). Governments should not be given carte blanche, even among electoral winners, as “a certain amount of rational distrust is necessary for political accountability in a participatory democracy” (Barber, 1983, p. 166). Some skepticism among winners is especially important as they also have more influence on those in offices than do losers.

Finally, a wide legitimacy gap is indicative of a polarized political system (Love & Carlin, 2014), which is associated with a number of negative outcomes, including gridlock, and decreased cooperation and tolerance (Persily, 2015). Thus, what is of most interest is not necessarily the absolute levels of perceived legitimacy among losers, but the legitimacy gap between winners and losers (Moehler, 2009).

Recognizing the importance of the legitimacy gap, a number of scholars have looked at factors that diminish or exacerbate the legitimacy gap. These factors are either measured at the individual level—such as strength of party attachment or ideology (Anderson et al., 2007)—or, more frequently, at the institution level—such as whether the system is majoritarian or consensual. Each of these characteristics “determine… how losers experience the exercise of power by a government they did not help elect—that is, the outcomes of government action citizens observe” (Anderson et al., 2007, p. 9). I propose a third variable that exists both at an organizational level but also at the system level: ideologically biased media and its country-level instantiation, political parallelism.

Media Bias and Democratic Attitudes

Biased media are maligned among political communication researchers.¹ The reemergence of overtly partisan news in the United States in the 1990s inspired a host of studies investigating the deleterious effects of these so-called echo chambers (Sunstein, 2009). That is, today’s media environment facilitates a person’s ability to avoid media that conflict with his or her political identity or attitudes. Past research indicates that this type of selectivity occurs at least among the politically interested (e.g., Goldman & Mutz, 2011; Prior, 2007), if not among the uninterested (Arceneaux & Johnson, 2013; Prior, 2007). Among those who do pay attention to partisan news, it serves as an “engine of polarization” (Mutz, 2006, p. 233), and polarization, in its various guises, has the potential to increase the legitimacy gap.

Since biased news increases ideological extremity (e.g., Levendusky, 2013b; Stroud, 2010; Sunstein, 2009, although see Arceneaux & Johnson, 2013; Prior, 2012), those who pay attention to such media are more likely to believe that the ideal points of the ruling cabinet are far from their own (if they are electoral losers) or closer than reality (if they are electoral winners) (Ahler, 2014; Levendusky & Malhotra, 2013; Sood & Iyengar, 2014). As voters are expected to be more satisfied with the output of government that is closer to their own ideal points (Curini, Jou, & Memoli, 2011; Huber & Powell, 2011), more polarized voters are expected to be less satisfied with government when in the opposition than more moderate voters.

Biased news also increases partisans’ affective polarization (Lelkes, Sood, & Iyengar, 2015; Levendusky, 2013b), their level of interparty animosity (Iyengar, Sood, & Lelkes, 2012). As biased media are marked by “efforts to provoke visceral responses (e.g., anger, righteousness, fear, moral indignation) from the audience through the use of overgeneralizations, sensationalism, misleading or partially inaccurate information, ad hominem attacks, and partial truths about opponents” (Sobieraj & Berry, 2011, p. 29), it’s easy to see how exposure could make a consumer hate the other side. In addition, mere exposure to partisan messages and symbols may strengthen partisan identity (Horwitz & Nir, 2014), which, in turn, heightens in-group favoritism and out-group animosity (Greene, 2012; Tajfel, Billig, Bundy, & Flament, 1971).

As opposed to the ideologically polarized citizen, the affectively polarized citizen may be judging the legitimacy of the government based on a purely emotional basis. Those who hate one party or another may distrust anything and everything associated with that party. Conversely, those who feel overly warm to one party may see anything related to its government through rose-colored glasses. This would be in line with motivated reasoning research that shows that citizens ignore or discount negative performance information about parties or politicians they like and positive information about parties or politicians they do not like (Lebo & Cassino, 2007). As government performance is believed to affect legitimacy (Gilley, 2006), biased media alter the way citizens judge the ingredients of legitimacy by impacting affective polarization and triggering motivated reason.

Biased media may also impact perceptions of legitimacy in a way that is, at least on its face, unrelated to polarization: losers may not perceive those in power as the rightful winners. Biased media often highlight the idea that the opposition is running unfair ads or cheating in other ways (Mutz, 2006). As Mutz puts it,

If one believes the other side won by running deceitful ads then it is easy to villainize the opposition and thus create more extreme perceptions of the consequences of one political choice over another. At this point, it is no longer about differing political philosophies; it is about right versus wrong, trust versus deceit good versus evil. (p. 278)

Media Bias in a Comparative Perspective: Political Parallelism

With few exceptions, recent investigations of the effect of partisan media on political attitudes and behaviors have focused on the United States. However, the American political information environment is fairly unique, and the generalizability of these findings outside the U.S. context is unclear. Biased news, although prevalent in the nineteenth century, is a new phenomenon in the United States. Partisan television shows emerged only in the post-broadcast era (Prior, 2007) and despite being a major focus of recent political communication research, the audience for these channels is quite small in comparison to that of mainstream local and network news (Prior, 2012).

Compare this to many other countries, where objective news is the exception rather than the rule. In many Southern and Eastern European countries, newspapers and television outlets are still explicitly linked to a political party. Newspapers and many television channels serve as mouthpieces for the major parties in Mediterranean states, such as Greece, Spain, or Italy, or Eastern European states. The notion of a disinterested journalist is much less applicable in these countries than it is in the United States; for instance, Russian “journalists who adapt to western standards of objectivity are often regarded as ‘robots’ by their more traditionally oriented colleagues” (Voltmer, 2011, p. 478). Journalists instead offer the news of the day in a much more interpretive, and often fiery, fashion.

Many countries in Northern Europe also have a history of overt bias, but, due to liberalization, there is now considerable variation across countries. In the early twentieth century, each major party in Denmark, for instance, printed its own newspaper; such explicit bias in Denmark is almost nonexistent today (Brüggemann, Engesser, Büchel, Humprecht, & Castro, 2014). While outlets in Northern Europe are rarely explicitly linked to specific parties, many outlets are still associated with an ideological point of view. In Germany, for instance, the Frankfurter Allgemeine tends to lean to the right, while Süddeutsche Zeitung, tends to lean left (Hallin & Mancini, 2004), and in the United Kingdom, while television news is decidedly centrist, newspapers like The Sun or the Mirror clearly mirror the point of view of the Conservative and Labour parties, respectively.

This degree to which media content in a country “reflect[s] distinct political orientations in their news and current affairs reporting” (Hallin & Mancini, 2004, p. 28) is commonly referred to as political parallelism. While the concept encompasses more than just the content of an outlet–other indicators include the explicit links between media outlets and actors and political actors (Allern & Blach-Ørsten, 2011; Hallin & Mancini, 2004; Seymour-Ure, 1974) and the audience’s political preferences (van Kempen, 2007)—the degree to which it may or not have an effect on attitudes or behavior is more likely a function of media bias than other aspects of parallelism, and has been operationalized in a variety of ways including the issue agendas of the outlet (e.g., Eilders, 1999) and differences in arguments and evaluations of political actors (e.g., Tresch, 2008).

Few studies have assessed the impact of this country-level variable on individual-level behavior. Goldman and Mutz (2011) find higher levels of political parallelism are associated with a lower likelihood of encountering dissimilar views; both Baek (2009) and van Kempen (2007) find that higher levels of political parallelism are associated with higher levels of political participation; and Horwitz and Nir (2014) find that political parallelism increases partisan attachment.

Since encountering partisan media is more likely in countries with higher levels of political parallelism, I hypothesize that the gap between winners’ and losers’ institutional trust would be greater in countries with relatively higher levels of political parallelism. However, parallelism on television can be unrelated to parallelism in newspapers (Goldman & Mutz, 2011; Hallin & Mancini, 2004; van Kempen, 2007), and each should be measured separately. Even within these countries, the probability that a person runs into biased media is contingent on whether or not they are exposed to it. Hence, the effect of newspaper political parallelism should be larger among those who read newspapers, and television political parallelism should be larger among those who watch television news.

To reiterate, this study explores two hypotheses:

• H1: The legitimacy gap is larger in more politically parallel countries than less politically parallel countries.

• H2: The relationship between political parallelism and the legitimacy gap is larger among those who pay more attention to the news.

In sum, past research indicates that partisan media should increase institutional trust and the perceived legitimacy of government among electoral winners and weaken it among electoral losers. The study described next tests whether this is indeed the case by comparing the legitimacy gap between winners and losers in 28 countries that vary in their degree of political parallelism. By doing so it ties together the fields of comparative research and media effects.

Data and Measures

To test, these hypotheses I utilize multiple sets of data. The individual-level data come from the three, four, and five rounds of the European Social Survey (ESS), which were fielded in 28 countries between 2006 and 2010. Most countries appear in each survey year, but some (e.g., Romania, Croatia, and Austria) only participated in one or two of the survey years. Strict random probability methods were used to sample inividuals within each country. Face-to-face surveys were conducted after a rigorous translation process of the questionnaire (to ensure cross-national comparability). The merged data set consisted of 125,528 complete cases–roughly 105,000 when non-voters are omitted. Sample sizes varied between, 1,681 (in Romania) to 7,618 (in Germany). The sample sizes by year for each country are available in the Supplemented Material. Multiple waves were combined, as the number of respondents within each country that do not watch any television news is fairly small (usually about 10% of the sample), and combining years increases non-watchers from roughly 100 per country to roughly 300.

Two sets of outcome measures were used to measure legitimacy. First, institutional trust was calculated as the the mean response to questions about R’s trust in various national institutions: parliament, the legal system, and the police (α = .91). Klingemann (1998) argues that this is a measure of trust in the political regime (rather than the specific authority in power), and fits closely with the notion that legitimacy is grounded in citizens’ trust in government to do what is right (Anderson & Singer, 2008; Easton, 1965; Hakhverdian & Mayne, 2012; Kees & Strömbäck, 2012; Klingemann, 1998). As a second measure of political legitimacy, I use a measure which asked, “And on the whole, how satisfied are you with the way democracy works in [country]?” These, and all other measures in the analysis, were recoded to lie between 0 and 1, with 1 indicating the highest possible level of political trust across regimes, and 0 the lowest possible level of trust.

Respondents were also sorted into winners or losers. I follow the standard coding in the literature (e.g., Anderson et al., 2007; Curini et al.. 2011; Hakhverdian & Mayne, 2012) and, utilizing past-vote choice, group respondents into those who voted for a party that was in the ruling coalition and those who voted for a party that was not.² I do not include non-voters in the analysis, as they cannot be easily grouped into either the ruling or the opposition category. Results are substantively the same if they are included and grouped with those not in power. Coding respondents who voted for a party that supports the ruling coalition but is not a formal member as a “winner” does not change the results. The survey was fielded in a number of countries before and after an election. Winner/loser status was predicated on the respondent’s interview date in these instances.

The country-level indicators of political parallelism come from the 2010 European Media Systems Survey (EMSS), which asked academic experts in 28 countries that were also in the ESS to evaluate individual television and newspaper outlets on a variety of dimensions. I use the sum of three variables as the media parallelism measure. For each outlet, experts were asked to indicate, on a 0-to-10 point scale: (a) the degree to which coverage was influenced by any party or parties, (b) the degree to which outlets advocated for specific policies, and (c) how well each outlet represents the arguments of all sides (reverse coded). Results are similar if only one of these measures is used rather than all three. EMSS reports interrater reliability scores above α = .85 for each of these variables across countries. For each outlet, average scores on each measure were weighted according to their audience share.³ This score represents the average political parallelism measure, with higher values indicative of a more biased media. Since it also incorporates audience shares, it is also indicative of the relative prevalence of partisan media. The three measures demonstrated satisfactory internal consistency .

I also utilize two other parallelism measures—political parallelism only among newspapers (newspaper parallelism) and political parallelism only among television stations (television parallelism)—test whether the effect on the legitimacy gap is larger for one medium than another. The correlation, between the two measures was fairly, strong although still indicative of discriminant validity (r = .33).⁴

These values, which appear in Figure 1 (in order of their level of overall parallelism), match Hallin and Mancini’s mostly qualitative analysis of political parallelism. Countries in Eastern Europe (e.g., Russia, Ukraine, and Bulgaria) and Southern Europe (e.g., Greece, Spain, and France) tend to exhibit high levels of political parallelism on both mediums, although political parallelism was lower on television than in newspapers. This is in line with the thesis that newspapers are geared toward elites in those countries, while the average (more centrist) citizen tends to get their news from television. Not surprisingly, Great Britain exhibits very high levels of political parallelism in newspapers, but is much more neutral on television. While newspapers in Northern European countries tend to exhibit moderate levels of political parallelism, television stations are much more neutral.

Figure 1. Newspaper, television, and cross-media political parallelism score by country.

The patterns that appear in Figure 1 highlight the possibility that a number of variables that predict parallelism also predict differences in political legitimacy. To deal with possible confounds, I include several country-level control variables in multivariate analyses that Anderson and Singer (2008) and others have shown to be related to the legitimacy gap:

First, political institutions can mitigate the winner/loser gap. More proportional systems alleviate the sting of losing through “fairer representation” (Anderson & Singer, 2008, p. 147) by allocating seats by vote share. Furthermore, more proportional systems tend to yield smaller shifts in policy outcomes, and, therefore, smaller shocks to the status quo. Therefore, I include both a commonly used index of electoral disproportionality (Gallagher, 1991) and an indicator of cabinet type (single–party majority cabinet versus coalition cabinet), both calculated for each country-year.

Second, older democracies and democracies with more experience with fair and honest elections experience a smaller legitimacy gap than newer democracies or countries with more corrupt elections (Moehler, 2009). As can be seen in Figure 1, more established democracies also tend to have lower levels of political parallelism; most of the younger democracies were part of the former Soviet Union. Hence, I include a variable that indicates whether a country is generally considered part of Central and Eastern Europe (CEE) and a variable that indicates whether a country is in Southern Europe.⁵ Third, countries rated higher in socioeconomic development tend to have a strongly professionalized press and a higher percentage of the population who tunes into the news. Socioeconomic development is also related to a smaller legitimacy gap (Anderson & Singer, 2008) in line with the argument that economic development is a prerequisite for pro-democratic attitudes (Lipset, 1959). Hence, I use the Human Development Index (HDI) for each country-year to control for these differences.

Fourth, I include a variable to indicate whether a country is federalist or not. When power is divided across levels, Anderson and Singer (2008) argue, the effect of losing in a national election is minimized as it is the ability of electoral winners to make sweeping policy changes by distributing power among different levels of government. Thus, I include a dichotomous federalism indicator variable.

Finally, I also include an index of a country’s level of corruption to control for country-level differences in legitimacy. The Corruption Perceptions Index (CPI), an expert survey administered by Transparency International, was rescaled to lie between 0 (highly corrupt) and 1 (not at all corrupt) (for details, see Lambsdorff, 2005).

Table 1 displays the summary statistics for the dependent variables, country level variables, and the two primary individual-level variables of interest. All variables are rescaled to lie between 0 and 1 by substracting the observation by each variable’s minimum score and dividing by difference between the minimum and maximum of that variable. Variables are dichotomous: the cabinet variable, the two region variables (Central or Eastern Europe and Southern Europe), the variable indicating whether the country is Federalist or not, and the variable indicating whether R voted for the ruling coalition or not. The distribution of the latter variable, which indicates that 45% of the sample voted for the ruling coalition in their country, closely mirrors that of Anderson and Colleagues (2007).

Table 1 Summary statistics

Statistic	N	Mean	SD
Media Parallelism	105,380	0.45	0.25
Newspaper Parallelism	105,380	0.42	0.26
Television Parallelism	105,380	0.48	0.23
Southern Europe	105,380	0.37	0.48
Central and Eastern Europe	105,380	0.29	0.23
Disproportionality	105,380	0.24	0.43
One-Party Cabinet	105,380	0.26	0.44
Human Development Index	105,380	0.57	0.24
Federalist	105,380	0.22	0.42
Corruption Perception Index	105,380	0.57	0.30
Satisfaction With Democracy	101,443	0.50	0.25
Institutional Trust	104,928	0.45	0.22
Total News Consumption	104,038	0.21	0.13
Newspaper Consumption	104,413	0.13	0.13
Television Consumption	104,816	0.29	0.19
Political Interest	105,032	0.50	0.29
Total Media Use	104,167	0.48	0.38
Winner/Loser	105,359	0.45	0.50
Age	104,956	0.42	0.20
Education in Years	104,281	0.22	0.08
Percentage Female	105,280	0.55	0.50

While I also explore the effects of political parallelism on the winner-loser gap across all respondents, my main hypotheses are conditional. The effect of the political parallelism should be larger among those who pay attention to the news. The interaction between parallelism and exposure is analagous to content-weighted media exposure measures used in many past studies (Bos, Van der Brug, & De Vreese, 2011; Jebril, Albæk, & De Vreese, 2013; Scheufele, 2000; Shehata & Strömbäck, 2013). To validate this measure, I also test whether it predicts political extremism, a well-known correlate of exposure to biased media. I fold an 11-point left-right measure around the midpoint and regress it on media parallelism, media exposure, and the interaction between the two. As expected, this interaction significantly predicts higher levels of political extremism (b = .25, SE = .13).

However, “a central issue in the study of media effects, particularly though not exclusively in cross-sectional data, is the problem of self-selection” (Soroka et al., 2013, p. 9). In the present case, a number of variables may predict both exposure to media as well as differences in institutional trust. For instance, more educated respondents are both more likely to read newspapers and exhibit higher levels of political trust. Hence, any analysis which considers the effect of news consumption must account for possible confounders.

I utilized coarsened exact matching (CEM; Iacus et al., 2011) to achieve balance on a variety of covariates among those who were exposed to more or less political news. That is, matching creates two groups that are identical on a variety of covariates known to predict news consumption, and therefore, approximates the experimental ideal of a randomized control group and a treatment group. Matching does not completely overcome the self-selection issue, but, compared to typical regression methods, is less dependent on modeling assumptions, and estimates are typically less statistically biased. CEM is an algorithm that “coarsens” or discretizes a continuous variable in order to create exact matches between “treatment” and “control’ within these strata of the variable. Units are weighted so that the number of “control” units in each stratum equals that of the number of “treatment” units. Units within strata that cannot be matched are weighted to 0. The ESS asked respondents to indicate the number of hours per day they spend watching political affairs television and reading about news political affairs in newspapers. I also average the two measures to form a more reliable indicator of political news consumption⁶ I dichotomize these measures by coding 1 if the respondent lies in the top half of the distribution of news consumption (“treatment”), and 0 if the respondent lies in the bottom half (“control”). Following Soroka and Colleagues (2013), political interest, education, age, age-squared, total media use, and gender are balanced across each “treatment” variable (exposure to news on television, news in newspapers, and news in either). See Supplemental Material for balance statistics.

Results

Trust in political institutions and satisfaction with democracy were regressed on the country-level variables and the individual-level winner/loser status indicator. To account for cluster-level unobserved heterogeneity, I allow the intercepts to vary between countries-year in a mixed-effects model. To determine whether the winner-loser gap changes as political parallelism increases, the interaction terms between winner/loser status and each political parallelism measure were included. A positive interaction term would indicate that the gap increases as political parallelism increases. Table 2 shows these interaction effects and the 95% confidence intervals.

Table 2 Media parallelism and the legitimacy gap

	Trust		Satisfaction w/Democracy
Intercept	−1.31^a	−1.34^a	−.85	−.91^a
	[−1.69; −.93]	[−1.74; −.95]	[−1.30; −.41]	[−1.37; −.44]
Southern Europe	−.03	−.03	−.02	−.02
	[−.10; .04]	[−.10; .04]	[−.10; .06]	[−.10; .06]
Central and Eastern Europe	.04	.03	.03	.02
	[−.05; .13]	[−.05; .12]	[−.07; .13]	[−.08; .12]
Human Development Index	1.95^a	1.94^a	1.21^a	1.19^a
	[1.57; 2.33]	[1.56; 2.32]	[.75; 1.66]	[.74; 1.64]
Disproportionality	.00	.00	.09*	.09*
	[−.02; .02]	[−.02; .02]	[.06; .11]	[.06; .11]
Federalism	−.01	−.01	.01	.01
	[−.09; .07]	[−.09; .07]	[−.08; .10]	[−.08; .09]
One−Party Cabinet	.01^a	.01^a	.03^a	.03^a
	[.01; .02]	[.01; .02]	[02; .04]	[.02; .03]
Corruption	.02	.02	.04^a	.04^a
	[−.01; .04]	[−.01; .05]	[01; .07]	[.01; .07]
Year: 2008	−.07^a	−.07^a	−.01^a	−.01^a
	[−.07; −.06]	[−.07; −.06]	[−.01; −.00]	[−.01; −.00]
Year: 2010	−.11^a	−.11^a	−.04^a	−.04^a
	[−.12; −.11]	[−.12; −.11]	[−.05; −.04]	[−.05; −.04]
Winner/Loser	.04^a	.03^a	.06^a	.05^a
	[.03; .04]	[.03; .04]	[.05; .06]	[.04; .06]
Media Parallelism	.05		.07
	[−.11; .21]		[−.11; .26]
Winner/Loser × Media Parallelism		.02^a		.03^a
	[.01; .03]		[.01; .04]
Newspaper Parallelism		−.02		−.02
		[−.13; .10]		[−.15; .11]
Television Parallelism		.11		.17
		[−.09; .31]		[−.06; .39]
Winner/Loser × Newspaper Parallelism		.01		.01
		[−.00; .02]		[−.00; .02]
Winner/Loser × Television Parallelism		.01^a		.02^a
		[.00; .02]		[.01; .04]
Akaike Information Criterion	−49797.77	−49781.47	−17202.82	−17189.87
Bayesian Information Criterion	−49654.37	−49618.91	−17059.91	−17027.91
Log likelihood	24913.89	24907.73	8616.41	8611.94
Num. obs.	104860	104860	101425	101425
Num. groups: country	28	28	28	28
Variance: country (Intercept)	.01	.01	.01	.01
Variance: residual	.04	.04	.05	.05

^a0 outside the confidence interval

On average, political parallelism was only weakly related to the winner-loser gap. Moving from the lowest to highest levels of media parallelism (the average of television and newspaper parallelism) is associated with a 2-point increase in political trust (b = .02, 95% CI = .01–.03; Table 2, Column 1). Results are similar if we separate that measure into its eomponent parts: Moving from the lowest to highest levels of newspaper parallelism is not associated with a change in political trust, and moving from the lowest to highest levels of newspaper parallelism is associated with a 1-point increase in political trust (b = .02, 95% CI = .00–.02; Table 2, Column 2). The relationship between the gap in satisfaction with democracy and parallelism is roughly the same. The gap in satisfaction with democracy is three points larger in the most parallel country than the least parallel country (b = .03, 95% CI = .01–.04; Table 2, Column 4). The gap is roughly .01 (95% CI = 00–.02; Table 2, Column 4) larger in countries with the most biased newspapers compared to the least biased newspapers, and .02 (95% CI = .01–04; Table 2, Column 4) larger in countries with the most biased television compared to the least biased television.

While the relationship between parallelism and the winner-loser gap on the full sample are of interest, the more stringent test is the aforementioned conditional effect: Political parallelism should have a larger effect on those who are exposed to the parallel media. The next set of analyses determine whether these interaction effects are larger among the top half of media consumers (total media, newspaper, and television, respectively) compared to the bottom half of media consumers.

As expected media parallelism was not associated with a larger gap in political trust among those in the lower half of total media consumption (b = .01, 95% CI = –.01 – .03; Table 3, Column 1, but was associated with a larger gap in political trust among those in the upper half of total media consumption (b = .04, 95% CI = .02 – .05; Table 3, Column 2). The three-way interaction testing the difference between Column 1 and Column 2 was b = .03 (95% CI = .01–.05; Table 3, Column 3). Similarly, political parallelism was larger among those in the upper half of media consumption (b = .06, 95% CI = .04–.07; Table 3, Column 4) than the lower half (b = .04, 95% CI= .02–.06; Table 3, Column 5). The confidence interval around the three-way interaction effect barely included zero (b = .02, 95% CI= –.01–.05, Table 3, Column 6).

Table 3 Media parallelism and the legitimacy gap, conditional on the amount of total, media consumption

	Bottom Half	Top Half	Both	Bottom Half	Top Half	Both
Winner/Loser	.04^a	.03^a	.04^a	.06^a	.05^a	.06^a
	[.04; .05]	[.02; .04]	[.04; .05]	[.05; .07]	[.04; .06]	[.05; .07]
Media Parallelism	−.24^a	−.28^a	−.24^a	−.20^a	−.26^a	−.21^a
	[−.38; −.09]	[−.44; −.13]	[−.39; −.09]	[−.36; −.05]	[−.43; −.10]	[−.37;−.04]
Winner/Loser × Medía Parallelism	.01	.04^a	.01	.04^a	.06^a	.04^a
	[−.01; .03]	[.02; .05]	[−.01; .03]	[.02; .06]	[.04; .07]	[.02; .06]
Consumption			.03^a			.03^a
			[.02; .04]			[.02; .04]
Winner/Loser × Consumption			−.02^a			−.01
			[−.03; −.00]			[−.02; −.00]
Media Parallelism × Consumption			−.05^a			−.05^a
			[−.06; −.03]			[−.07; −.03]
Winner/Loser × Media Parallelism × Consumption			.03^a			.02
			[.01; .05]			[−.01; −.05]
Akaike Information Criterion	−1456.71	−23604.34	−25074.06	10718.91	−9492.30	1278.01
Bayesian Information Criterion	−1406.70	−23553.11	−24982.70	10768.92	−9441.07	1369.37
Log likelihood	734.36	11808.17	12547.03	−5353.45	4752.15	−629.01
Num. obs.	30792	37763	68555	30792	37763	68555
Num. groups: country	28	28	28	28	28	28
Variance: country (Intercept)	.01	.01	.01	m	.01	.01
Variance: Residual	.03	.03	.03	.05	.05	.05

^a0 outside the 95% confidence interval. Covariates included in analysis, but omitted from table for presentational purposes.

The results are similar if we split up the measures into newspaper parallelism and newspaper consumption. Newspaper parallelism was not related to the size of the winner-loser gap in political trust among those in the bottom half of newspaper, consumption (b = –.02, 95% CI = –.03–.00; Table 4, Colunm 1), but was among those in the upper half of newspaper consumption (b = .04, 95% CI= .03–.05; Table 4, Column 7). The three-way interaction testing the difference in effects was b = .06 (95% CI = .03–.08). Similarly, the gap in satisfaction with democracy was not related to newspaper parallelism among those who read newspapers the least often (b = –.01, 95% CI= –.03–.01; Table 4), but was related to newspaper parallelism among those who read newspapers the most often (b = .04, 95% CI = .02–.04; Table 4, Column 5). The confidence interval around the three-way interaction testing the difference in effects did not include zero (b = .05, 95% CI = .02 to .08; Table 4 Column 6).

Table 4 Newspaper parallelism and the legitimacy gap, conditional on the amount of total newspaper

	Bottom Half	Top Half	Both	Bottom Half	Top Half	Both
Winner/Loser	.05^a	.03^a	.06^a	.09^a	.06^a	.09^a
	[.05; .06]	[.02; .04]	[.05; .06]	[.08; .10]	[.05; .07]	[.08; .10]
Newspaper Parallelism	−.10	−.13	−.09	−.05	−.10	−.05
	[−.26; .06]	[−.29; .04]	[−.26; .07]	[−.22; .12]	[−.27; .07]	[−.22; .12]
Winner/Loser × Newspaper Parallelism	−.02	.04^a	−.02	−.01	.04^a	−.01
	[−.03; .00]	[.03; .05]	[−.03; .00]	[−.03; .01]	[.02; .05]	[−.03; .01]
Consumption			.04^a			.04^a
			[.04; .05]			[.03; .05]
Winner/Loser × Consumption			−.03^a			−.03^a
			[−.04; -.01]			[−.04; .02]
Newspaper Parallelism × Consumption			−.04^a			−.05^a
			[−.05; −.02]			[−.06; −.03]
Winner/Loser × Newspaper Parallelism × Consumption			.06^a			.05^a
			[.03; .08]			[.02; .08]
Akaike Information Criterion	−2131.02	−29362.27	−31419.19	7898.13	−12173.02	−4073.44
Bayesian Information Criterion	−2082.43	−29309.87	−31327.61	7946.71	−12120.62	−3981.86
Log likelihood	1071.51	14687.14	15719.59	−3943.06	6092.51	2046.72
Num. obs.	24276	45850	70126	24276	45850	70126
Num. groups: country	28	28	28	28	28	28
Variance: country (Intercept)	.01	.01	.01	.01	.01	.01
Variance: residual	.03	.03	.03	.05	.04	.05

^a0 outside the 95%, confidence interval. Covariates included in analysis, but omitted from table for presentational purposes.

Television parallelism yielded similar results. The winner-loser gap was not related to parallelism among those in the bottom half of television consumption (b = –.00, CI = –.02–.02; Table 5, Column 1), but was among those in the upper half of television consumption (b = .03, CI= .02–.05; Table 5, Column 2). Finally, the gap in satisfaction with democracy was smaller among the lower half of television consumers (b = .03, 95% CI = .00–.05; Table 5, Column 4) than among the upper half of television consumers (b = .07, 95% CI = .05–.09; Table 5, Column 5). The confidence interval around the three-way interaction assessing the difference in-between Column 4 and Column 5 did not include zero (b = .03, 95% CI = .00–.06; Tabel 5, Column 6).

Table 5 Television parallelism and the legitimacy gap, conditional on the amount of total television consumption

	Bottom Half	Top Half	Both	Bottom Half	Top Half	Both
Intercept	.56^a	.58^a	.56^a	.58^a	.61^a	.59^a
	[.48; .65]	[.49; .67]	[.48; .65]	[.49; .67]	[.51; .70]	[.49; .68]
Winner/Loser	.05^a	.03^a	.05^a	.07^a	.04^a	.06^a
	[.04; .06]	[.02; .04]	[.04; .06]	[.05; .08]	[.03; .05]	[.05; .08]
Television Parallelism	−.28^a	−.32^a	−.28^a	−.25^a	−.31^a	−.26^a
	[−.43; −.12]	[−.49; −.15]	[−.44; −.12]	[−.42; −.09]	[−.49; −.13]	[−.43; −.09]
Winner/Loser × Television Parallelism	−.00	.03^a	.00	.03^a	.07^a	.03^a
	[−.02; .02]	[.02; .05]	[−.02; .02]	[.00; .05]	[.05; .09]	[.01; .06]
Consumption			.02^a			.02^a
			[.01; .03]			[.01; .03]
Winner/Loser × Consumption			−.02^a			−.02^a
			[−.03; −.00]			[−.04; −.01]
Newspaper Television × Consumption			.04^a			.05^a
			[−.06; −.02]			[−.07; −.02]
Winner/Loser × Television Parallelism × Consumption			.03^a			.03^a
			[.00; .05]			[.00; .06]
Akaike Information Criterion	−2414.67	−26172.19	−28600.29	7573.66	−10216.06	−2617.33
Bayesian Information Criterion	−2365.63	−26120.16	−28508.82	7622.70	−10164.02	−2525.86
Log likelihood	1213.33	13092.10	14310.14	−3780.83	5114.03	1318.66
Num. obs.	26175	43157	69332	26175	43157	69332
Num. groups: country	28	28	28	28	28	28
Variance: country (Intercept)	.01	.01	.01	.01	.01	.01
Variance: residual	.03	.03	.03	.05	.05	.05

^a0 outside the 95% confidence interval. Covariates included in analysis, but omitted from table for presentational purposes.

Robustness Checks

In order to test the robustness of these findings, I employ a number of different strategies the results of which appear in the Supplemental Material:

• To ensure that the results are not driven by one’country, I perform a jackknife test (van Kempen, 2007). That is, I again test whether the three-way interactions reported earlier substantially change when any of the countries are omitted from the analysis.

• Rather than compare the the lower half and upper half of media consumers, I compare the bottom third to the upper third.

• I utilize nearest neighbor propensity score matching. In addition, for expositorv purposes, I also show results without any matching and show the results using the full continuous measure and without matching.

• As partisan media are more likely when political elites are polarized (Ladd, 2011), and elite polarization is linked to citizen polarization (Layman, Carsey, & Horowitz, 2006) the effect of partisan media may be spurious. While sorting out the endogeneity is not possible with cross-sectional data, some comfort can be gained if the effect of partisan media on the legitimacy gap persists in controlling for the effect of elite polarization. I calculated party polarization scores based on Dalton’s index (Dalton, 2008) using the the second and third module of the CSES, and use the most recent score from those modules. However, the CSES data are available in only 25 of the countries that I include in the main data set. Regardless, the results are substantively equivalent.

Discussion

Scholars have long stressed not only the overall importance of citizens’ perceptions of democratic legitimacy, but also the gap in legitimacy between winners and losers. Political scientists have, primarily, focused on the institutional-level variables that exacerbate or minimize this gap. The current study points to another factor—the media—that may be equally important in explaining the winner-loser gap. In fact, the structure of the political information may be a more proximal factor that influences perceptions of legitimacy, as it filters citizens’ experiences with the institutional factors.

Specifically, the degree to which the newspapers and television parallel the political environment mediates the impact of winning or losing an election on political trust and satisfaction with democratic processes. The overall degree of political parallelism is associated with changes in the legitimacy gap, but the clearest evidence that political parallelism affects perceptions of legitimacy emerges when we separate out those who consume more media from those who consume less media. In general, the winner-loser gap was not related to parallelism among those in the bottom half of news consumption, but was among those in the top half of news consumption.

The effects of political parallelism on perceived legitimacy are not large, but media effects rarely are (Valkenburg & Peter, 2013). However, this was a conservative test of the hypotheses. First, the exposure measure is quite blunt. We do not know whether those who read newspapers read more extreme or more centrist newspapers, and we assume that higher levels of partisan media availability are indicative of more partisan media consumption. Furthermore, while widely used, the ESS measures are still self-reports, and are, therefore, problematic. Second, we could expect substantial variance in the effects of media exposure within countries within more parallel countries, with larger media effects coming from extreme papers than centrist papers. Finally, the data were limited to Europe, where (a) the winner-loser gap is quite small compared to other places, and (b) people, especially in Western Europe, tend to be politically informed. Both these conditions constrain the likelihood of finding media effects. However, it should be noted that these small media effects are as large as the effects of the institutional variables found in the literature (e.g., Anderson et al., 2007; Moehler, 2009; Moehler & Lindberg, 2009; Nadeau & Biais, 1993).

Some concerns over measurement remain and cannot be dismissed. First, support for the winning party may be endogenous to trust. A best-case scenario to overcome these issues would be to identify a variable that predicts supporting the winner (versus loser) in a model predicting political trust but not with the error term in that model. Identifying such an instrument is difficult enough in single-country studies that correlate political attitudes, and exponentially more difficult in a comparative case. Second, the measure of exposure does not necessarily indicate that citizens living in highly parallel countries are necessarily consuming partisan news. Because I am using an audience-weighted structural measure of the media environment and self-reported exposure to the general environment, I can only make the assumption that those living in more parallel countries are more likely to be exposed to partisan news.

While this study takes the role of the media seriously in explaining the legitimacy gap, political parallelism is only one aspect of the political information environment. Other aspects of a country’s media may be equally or more important. For instance, recent work has emphasized the importance of broadcast versus commercial media in predicting citizen attitudes, knowledge, and behavior (e.g., Iyengar et al., 2010; Kees & Strömbäck, 2012; Soroka et al., 2013; Tworzecki & Semetko, 2012). An independent public broadcaster is more likely to provide balanced coverage of political actors than a commercial one, which may be more affected by the views of its audience or its owners. In addition, a public broadcaster is associated with higher leveks of political knowledge (Iyengar et al., 2010; Soroka et al., 2013) and trust (Kees & Strömbäck, 2012). Broadcast media, therefore, may diminish the legitimacy gap.

Finally, these analyses do not tell us if political parallelism makes losers less trusting, winners more trusting, or both. This important question can only be answered with longitudinal pre-election/post-election data. This study also does not delve into ways in which political parallelism interacts with other country-level variables to affect the winner-loser gap. As the Arceneaux, Johnson, Lindstädt, and Vander Wielen (2013) and Clinton and Enamorado, (2014) studies demonstrate, partisan news seems to affect elite behavior, which then may affect citizen attitudes and behavior. Finally, several potential mechanisms that link political parallelism to political legitimacy were previously mentioned: affective polarization, ideological polarization, and perceptions of malfeasance. It is still unclear if any or all these mechanisms are at work, and this offers a potential avenue for future research.

The functioning of a democracy is dependent on perceptions of legitimacy. While the majority of past work has focused on ways in which political institutions affect trust and legitimacy, the present work demonstrates that we also need to consider the ways in which the political information environment underpins these perceptions.

Acknowledgments

I thank Kevin Arceneaux, Bert Bakker, Katjana Gattermann, Rebecca Weiss, Erica Weitz, Sean Westwood, anonymous reviewers, and seminar participants at the University of Amsterdam and the University of Southern Denmark.

Supplemental Material

Supplemental data for this article can be accessed on the publisher’s website.

Additional information

Notes on contributors

Yphtach Lelkes

Yphtach Lelkes is Assistant Professor of Political Communication, Amsterdam School of Communication Research, University of Amsterdam.

Notes

1. Although Dilliplane (2011) finds that exposure to partisan media increases political participation, which is, on its face, a benefit of biased media.

2. Another possible approach is going not by vote choice but by which party the respondent feels closest to (Moehler & Lindberg, 2009). However, this question was not asked in several countries in the ESS.

3. See Popescu, Gosselin, and Pereira (2011) for a complete discussion of the weighting procedure and other technical details of the EMSS.

4. For the purpose of validation, the EMSS political parallelism scores were compared to those calculated by van Kempen (2007). Although parallelism scores were only 14 countries in the van Kempen (2007) study overlapped with the EMSS database, the results are reassuring. The correlation between the media parallelism measure used in this study and van Kempen’s media parallelism measure was .53.

5. Results are substantively similar if I include the Freedom House indicator of whether a country is “free” or not.

6. While self-reported media use measures are of dubious validity, a recent analysis found, with regards to a similar measure, that “increases in this self-reported measure reflect increases in the actual amount of news to which respondents were exposed” (LaCour & Vavreck, 2014, p. 411). Soroka and Colleagues (2013) also show a strong .correlation between, similar, media use rneasures and political knowledge and while the ESS does not contain any political knowledge measures, total news consumption is strongly correlated with political interest (r = .38). In addition, measures like these are frequently used in the political communication literature (e.g., Boomgaarden & Vliegenthart, 2009; Hopmann, Vliegenthart, De Vreese, & Albæk, 2010; Jerit, Barabas, & Bolsen, 2006).