Media-driven polarization. Evidence from the US

The authors use US data on media coverage of politics and individual survey data to document that citizens exposed to more politicized newspapers have more extreme political preferences. This polarization effect of media is mainly driven by individuals who harbor liberal opinions reading more newspapers, as opposed to individuals endorsing rather conservative positions. More politicized media also reinforce other aspects of citizens’ political sophistication such as political knowledge. This enhanced political sophistication materializes in observable involvement in politics, measured by campaign contributions. JEL K4 H0


Introduction
Most democracies have never been as politically polarized as they are today. 1 Holding polar political opinions bears important socio-economic consequences, however, since in such contexts political decisions are likely to deeply divide the population and to bring along important economic costs. One emblematic such consequence has been the recent 35 days government shutdown in the U.S. (December 22, 2018to January 25, 2019, the longest in the country's history, whose root cause lied in a disagreement over the funding of the U.S-Mexico border wall. Delving into the determinants of political polarization, scholars have pointed at socio-economic factors (e.g. Grechyna 2016). To complement these findings, a growing literature in political economy has focused on the specific role of media, but the results are mixed. Campante and Hojman (2013) and Melki and Pickering (2014), for example, provide evidence showing that, historically, the increasing media penetration over time reduced polarization in US counties and OECD countries, respectively. However, according to Prior (2013), the emergence of more partisan media may have contributed to political polarization leading Americans to support more partisan policies or candidates.
Part of the lack of convergence in the findings can be attributed to the differential effects of different media sources on political attitudes and polarization: the media differ along several dimensions, including their informational content, the intensity of exposition, as well as the type of public that consumes them. The Nixon-Kennedy 1960 presidential debate constitutes a famous example of the strikingly different impact on voters' perception of TV and radio, respectively, with polls revealing Kennedy to have convinced relatively more TV viewers as compared to radio listeners (e.g. Druckman 2003). With the evolution of media technology, the attention of the scholarship therefore turned to these new sources of information. The introduction of the TV modified the playground in politics by attracting away some attention from the more informative traditional means of information, i.e. newspapers and radio, by producing a sharp decline in turnout in the 1950s (Gentzkow 2006), and also by contributing to the recent rise in political polarization in the U.S. (Martin and Yurukoglu 2017). With the quick rise of internet penetration and the development of social media, researchers explored whether trends in political polarization could be attributed to this new type of media. State of the art research, however, fails to establish a definite effect of internet and social media on political polarization (e.g. Halberstam and Knight 2016, Boxell et al. 2017, Lelkes et al. 2017. It has been recognized that newspapers occupy a central role in informing citizens, and yet their impact on political polarization remains unexplored. Kennedy and Prat (2017) show in a study on 36 countries that 40% of the surveyed subjects get informed by reading newspapers. Combining this finding with the strong political content of newspapers (Campante and Do 2014), one realizes the importance of exploring the potential causal effect of newspapers on political polarization.
In this paper we analyze the role of newspapers on citizens' political positions, and thereby on the ensuing political polarization, while accounting for the political content of news. Moreover, we study the mechanisms through which media-driven polarization takes place as well as potential implications. Our measure of media exposition is provided by Campante and Do (2014) who propose a measure of state media coverage of politics based on nearly four thousand newspapers all over the United States. We explore the effect of newspaper coverage combined with individual survey data from the 2008 wave of the American National Election Studies (ANES) on whether surveyed individuals have an interest in politics and read news. Our results support that in the states with greater media coverage of politics, an increase in the citizens' exposure to newspapers is associated with more extreme political views. In support of a causal effect running from the media to political opinions, we propose a placebo test by replicating the estimation with political opinions' data from 2000, but by using the 2008 measure of state media coverage as an explanatory variable. Reassuringly, the placebo estimation yields statistically non-significant estimates.
Interestingly, we uncover an asymmetric effect of media coverage on political opinions since our results indicate that media coverage induces more extreme political views only on individuals harboring liberal opinions. By further exploring the potential channel explaining this asymmetry, we uncover that the effect of newspaper exposition is conditional on individuals reading newspapers, with conservatives relying less on newspapers as a source of information. Our central result therefore suggests that politicized information polarizes political opinions. This could be interpreted as a confirmation of previous findings that news spread quickly in networks of the same political side (Halberstam and Knight 2016), eventually producing more polarized political attitudes.
Consistently, we also find that more politicized media improve individuals' political sophistication as measured as political knowledge regarding the U.S. political life. 2 Lastly, we demonstrate that an additional positive implication of a high media coverage of politics is a stronger involvement of citizens in politics, as measured as campaign contributions.

Data and Empirical Strategy
Our measure of media coverage of politics comes from Campante and Do (2004). They looked at newspapers whose print edition content is available online and searchable on the website NewsLibrary.com covering nearly four thousand outlets all over the United States in 2008. They searched for the names of each state's then current governors as well as, alternatively, for terms such as state government, state budget, or state elections, where 'state' refers to the name of each state. They computed the state-level measure of political coverage by taking the first principal component of the four search terms for each newspaper (adjusted by size), and performed a weighted sum of this measure over all newspapers. 3 They provide this index for 47 states as of 2008. 4 The mean value of the variable weighted by the geographical concentration of the newspapers is −0.21 and its standard deviation is 0.96. As 2 This result complements Snyder and Strömberg (2010) finding that voters living in areas where the press covers their U.S. House representative more are more likely to recall their representative's name and more able to rate him or her. 3 They used two alternative sets of weights: the circulation of each newspaper in the state and that circulation weighted by its geographical concentration. The latter, our preferred measure, put more weight on circulation closer to the capital. 4 Rhode Island is excluded from the sample since, as underlined by Campante and Do, it is a outlier − about 5 standard deviations greater than the state with the next largest measure. Campante and Do explains that "this is because there is one newspaper, the Providence Journal, that far outstrips the circulation of all other RI-based newspapers in the sample, shown in Figure 1, the lowest political coverage is found in Delaware (−2.85) and the highest in Virginia (1.99).
Our aim is to analyze the effect of media coverage on citizens' political preferences, under the presumption that if citizens from both sides adopt on average more (less) extreme ideological positions, the ensuing political polarization should increase (decrease). We use the 2008 wave of the American National Election Studies (ANES), which provides high quality survey data pertinent to important questions about vote choice, public opinion and related matters in the context of the 2008 United States presidential election opposing the Democratic candidate, Barack Obama, to the Republican candidate, John McCain. We construct a set of individual variables available for up to 2079 respondents.
Our measure of political attitudes relies on a self-placement of individuals in the 2008 ANES on a 7-point liberal-conservative scale ranging from "extremely liberal" to "extremely conservative". Our main variable of interest, Extreme Position, is a dummy coded 1 for respondents declaring to be at one of the extremes of the spectrum; i.e. "extremely liberal" (1 point), "liberal" (2 This newspaper had a very large measure of coverage of state politics, and is idiosyncratically driving the state-level measure". points), "conservative" (6 points) or "extremely conservative" (7 points). 5 In our sample, 42.2% of the total respondents can be considered to have extreme ideological positions. Regarding the variation across states, as shown in Figure 1, the state with the lowest proportion of respondents endorsing extreme positions is Illinois with less than 20% and the state with the highest proportion is Tennessee with more than 60%.
As an alternative measure of political attitude, we use a measure of preferences for political parties. In the 2008 ANES, respondents are asked to rank separately the Democratic and the Republican parties on a 11-point scale ranging from 0 ("strongly dislike Dem./Rep. party") to 10 ("strongly like Dem./Rep. party"). 6 In order to capture extreme political preferences, we then construct the variable Extreme Party Pre f erence which takes the value of 1 for respondents declaring to strongly like (9 or 10 points) or strongly dislike (0 or 1 point) either party. This variable is positively but weakly correlated with (.23) Extreme Position suggesting that they measure different dimensions of respondents' preferences.
Among the other outcomes of interest, Campaign contribution is a dummy coded 1 if the respondent states that she "contributed money to specific candidate campaign", referring to the 2008 presidential election. 7 Party contribution is a dummy for wether the respondent states that he/she "gave money to a political party during this (2008) election year." We also use variables evaluating political knowledge, namely dummies for whether the respondent knows the party with the most members in the House of Representatives before the election (House), in Senate (Senate) 8 and whether he/she knows the name of the Speaker of the House (Speaker). 9 Given that respondents are not homogeneously exposed to the press, we construct different variables of exposure which are equally based on the 2008 ANES. The first one, Campaign Interest is a measure of the respondent's interest in politics in general, ranging from 1 to 5 with higher values capturing higher levels of interest. 10 A second, more specific measure of exposure to newspapers, Read paper, is coded 1 if the respondent reports that he/she has "read about the (2008) presidential campaign in newspapers". Alternatively, Attention paper reports the level of attention paid by the respondent to "news about national politics in printed newspapers." We also construct a variable of exposure to other media with TV news capturing the level of attention paid to "news about national politics on TV". Finally, our analysis controls for a set of individual characteristics, which are age, income, education level, gender, household size, residence in urban area, and time between interview and election, as of 2008. 5 The intermediate values correspond to "slightly liberal" (3 points), "Moderate; middle of the road" (4 points) and "slightly conservative" (5 points). The exact question is "We hear a lot of talk these days about liberals and conservatives.
Here is a seven-point scale on which the political views that people might hold are arranged from extremely liberal to extremely conservative. Where would you place YOURSELF on this scale, or haven't you thought much about this?" 6 The exact question is "I'd like to know what you think about our political parties. After I read the name of a political party, please rate it on a scale from 0 to 10, where 0 means you strongly dislike that party and 10 means that you strongly like that party". 7 The exact question is "During an election year people are often asked to make a contribution to support campaigns. Did you give money to an INDIVIDUAL CANDIDATE running for public office?." 8 The question is "Do you happen to know which party had the most members in the House of Representatives in Washington/in the U.S. Senate BEFORE the election (this/last) month?" 9 The question is "The first name is NANCY PELOSI. What job or political office does she NOW hold?" 10 The question is "How interested are you in information about what's going on in government and politics?" www.economics-ejournal.org To analyze the effect of individual characteristics (exposure to media) conditional on a variable aggregated at a higher level (state media), we follow Facchini and Mayda (2009) and estimate the following probit model: where Φ(.) represents the cumulative distribution function of a standard normal, Y i captures the individual's ideological (party) position, monetary contribution or political knowledge, Interest i the individual's interest in following campaign or media exposure, Media s the state media coverage of politics in the individual's state of residence, X i is a vector of individual controls, ψ s state dummies to control for additive state-specific unobserved effects, and standard errors adjusted for clustering at the state level. Figure 1 depicts the scatter plot of the state-average "polarization", measured as the share of individuals with extreme ideological positions, versus the state media coverage of politics. Ideological polarization is higher in states with a higher media coverage of politics, which − if taken at face value − goes against prior arguments that higher media penetration decreases political polarization (e.g. Hojman 2013, Melki andPickering 2014).

Ideological Position and Media Coverage
To provide quantitative support of such a possibility, we then turn to the individual data analysis and first estimate equation (1) with Extreme Position as a dependent variable in Table 1. Column (1) shows that the positive impact of the respondents' interest in politics on their probability to adopt extreme (liberal or conservative) positions is magnified in states where the media coverage of politics is stronger. Figure 2 plots the marginal effect of individuals' interest level on the probability of endorsing extreme positions, for given media exposure. The observed effect is economically significant. Exposure to media in the states with the lowest media political coverage is not associated with more extreme positions, while media exposure increases the probability of adopting extreme positions by more than 40% in the states with the largest coverage, such as as Virginia and California. Column (2) shows that this relationship holds when using a more specific measure of exposure to newspapers, which is whether the respondent has read about campaign in newspapers.
At this stage we cannot exclude a causal effect running from the individuals' opinion to the media coverage of politics, or even perhaps the non-inclusion of some omitted variable. To convince the reader that the link is indeed causal, we propose in column (3) a placebo estimation by replicating column (1) with all the variables as of 2000 (from the 2000 wave of ANES) instead of 2008, except for the media variable still measured in 2008. We observe that the interacted variable is not statistically significant any more, suggesting that the media coverage in 2008 does not account for the past polarization of 2000. 11 Interestingly, we provide evidence that the polarization effect of media is mostly driven by liberal respondents. Indeed, column (4) replicates the specification of column (1) by restricting the sample to respondents positioning themselves as extremely liberal, liberal, slightly liberal or moderate, while column (5) restricts the sample to those respondents positioning themselves as extremely conservative, conservative, slightly conservative or moderate. Upon inspection of Table 1, we observe that the level of "campaign interest" is associated with more extreme positions in both cases. On the other hand, we observe that the effect of campaign interest conditional on media coverage is present only for the liberal sample. This finding indirectly suggests that liberals' political attitudes are more elastic to the media coverage of politics, contrasting therefore previous findings on the elasticity of conservatives to slanted TV news (Martin and Yurukoglu 2017). Columns (6) and (7) confirm this asymmetric pattern with our alternative dependent variable Extreme Party Pre f erence. This last set of estimations brings further support to the finding that the polarization effect of media is due to liberal individuals adopting more extreme liberal positions in the presence of highly politicised media while conservative individuals remain insensitive to the newspapers' activity. In the next section we propose a rationale for this finding.

Asymmetric Media Exposure
A possible explanation for the asymmetric effect of media exposure on liberals' and conservatives' political attitudes rests on the potential asymmetric exposure to media of these two types of individuals. Liberals could be more sensitive to newspapers' coverage of politics because they read more newspapers when conservatives are more exposed to TV news; an explanation that would be consistent with Martin and Yurukoglu (2017). We investigate this possibility in Table  2 by regressing different measures of exposure to media on an ordered measure of conservative ideology ranging from 4 (middle of the road) to 7 (strongly conservative) for the sample of rather conservative respondents (from 4 to 7) in the odd columns. In the even columns we replicate this exercise for the sample of rather liberal respondents.
From Columns (1) and (2), we deduce that more extreme ideology correlates positively with the respondents' interest level in politics, for both liberals and conservatives. Columns (3) to (6), however, provide some support that more liberal respondents do pay more attention to newspapers than moderate voters, while more conservative individuals tend to read less newspapers than mod-erate voters. This result could explain the asymmetric effect of media coverage on the political attitudes of liberals and conservatives, as observed in Table 1. Indeed, this evidence would suggest that conservatives do not endorse more extreme positions in the presence of more politicised media because they pay less attention to newspapers. On the other hand, liberal individuals who are shown to read more newspapers, are then confronted to more politicised media, and this in turn could radicalize their political attitudes.
One could object that conservatives may acquire relatively more information from alternative media sources, the most obvious candidate being TV. In such an instance, the exposure to politicised media could be different than the one assumed in our analysis, thereby jeopardizing our interpretations. The results of columns (7) and (8) show that respondents with more extreme positions pay more attention to news about national politics on TV in both the conservative and the liberal sample. 12 This suggests that TV is not a information channel specific to the conservatives, as opposed to the newspapers which are principally influencing liberals.

Political Knowledge and Money Contribution
Consistent with the evidence that politicized media radicalise political preferences (at least for the liberals), we next show that they also tend to reinforce other aspects of the citizens' political sophistication, i.e. the capacity for a person to acquire knowledge on politics, assimilate information and form political views. Indeed, our analysis uncovers that exposure to politicized media leads to improved political knowledge. The estimations of Table 3 show that exposure to media is associated with a higher probability that an individual recognizes the political colour of the majority in the House of Representatives (column 1), in the Senate (column 2) and the name of the Speaker of the House (column 3), in the states with a better coverage of politics. Dividing the sample into conservatives and liberals, Table 4 shows that this effect holds for both sides of the political spectrum. Melki and Pickering (2014) consider two competitive hypotheses regarding the effect of the media on individuals' political preferences. According to the "benign" view, greater media access produces a better-informed electorate, which may facilitate political consensus (for instance on whether a specific policy is efficient or not), which should in turn reduce polarization. On the contrary, the "malign" view supports that individual political preferences become more extreme when people are confronted to news. This is a self-selection mechanism whereby out of the several existing sources of information individuals select the ones more in line with their prior beliefs, and that eventually radicalizes their priors. Although our results do not provide tangible evidence of such a self-selection mechanism, they are nevertheless going against the benign view as they support that more politicized media increase citizens' political knowledge while radicalizing the political preferences of a part of the electorate, thus contributing to a more polarized electorate.
Finally we provide evidence according to which the enhanced political sophistication entailed by more politicized media materializes in observable involvement in politics. Indeed, using two measures of monetary contributions to politics as the dependent variables in Table 5, we find that 12 In an unreported regression that further confirms these results, we replicated this regression with a variable measuring the attention level given to TV news on Presidential campaign.
www.economics-ejournal.org the positive impact of the respondent's interest in politics on the probability to contribute to a specific candidate during the campaign (columns 1 to 3) or to a political party (columns 4 to 6) is magnified in the states with a higher media coverage of politics. When splitting the sample into rather conservative and rather liberal respondents, we observe this effect to hold for both types of individuals. This result complements earlier work on the U.S. showing that the government invests disproportionally more on groups that have more access to the media, since we show that citizens also participate more to politics when exposed to media.

Conclusion
Analysts of politics lament that political polarization has been increasing in the US, and in many other countries around the world. In this article we provide evidence that exposure to politicized media may lead to a radicalization of individual political views, thereby producing more political polarization. Interestingly, our results also suggest that political polarization may go hand in hand with a higher political sophistication of citizens, more informed and interested in politics. More generally, we show that the media-driven polarization is accompanied by other potentially beneficial evolutions for democracies, which are the improved political knowledge and higher political participation of citizens.