Loud and Clear: Effects of Homogenous and Extreme Partisan Media Diets

Douglas M. Allen and Devra C. Moehler

The Annenberg School for Communication University of Pennsylvania

[Please do not cite or circulate without permission from the authors]

This version: August 20, 2013

Keywords: Public Opinion; Media Effects (Other); Political Psychology; Participation; Quantitative - Survey

The explosion of cable television and talk radio programming allows individuals to selectively consume opinionated media from only one side of the political spectrum. Observers worry that media fragmentation along partisan lines polarizes the citizenry. However, unbalanced media consumption may also mobilize individuals to participate in the electoral process. We test the effects of exposure to 73 news and entertainment programs on individuals’ issue polarization and campaign participation using the 2008 National Annenberg Election Survey. We construct a measure of the homogeneity and ideological extremity of a person’s total media diet. Within- subjects and matching analyses indicate that lopsided partisan media diets increase campaign participation, but not polarization. Consumption of loud and clear partisan programming may enhance participatory democracy without sacrificing deliberative democracy.

Abstract: 122 words Manuscript: 8415 words

Acknowledgements: We are deeply indebted to the guidance and thoughtful feedback provided by Susanna Dilliplane, Ted Brader, Matt Levendusky, Marc Meredith, and Andrew Therriault.

Correspondence concerning this article should be addressed to Douglas Allen, Annenberg School for Communication, 3620 Walnut St., Philadelphia, PA, 19104. E-mail: [email protected]

LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 1

The recent proliferation of media sources provides individuals with more choice than ever before. Increased competition for audiences motivates targeted entertainment programs and partisan news shows that eschew journalistic norms of “objectivity” and balanced reporting in favor of opinionated commentary by openly partisan hosts (Groseclose & Milyo 2005;

Mullainathan & Shleifer 2005). These changes in the information environment allow citizens to select a media diet that contains more ideologically homogenous and extreme programming than was possible when more temperate network channels dominated the airwaves. However, selective exposure to likeminded news programming remains uneven and incomplete. Audience ratings show that network news programs still attract much larger audiences than even the most popular cable news shows (Arceneaux & Johnson 2008; Johnson & Arceneaux 2011; Webster

2005), and many cable news consumers report that they watch programming that ostensibly conflicts with their political views.i Nonetheless, a growing body of evidence indicates that a sizeable portion of the population now chooses to consume only biased programming that is compatible with their political views (Iyengar & Hahn 2009; Iyengar, Hahn, Krosnick, & Walker

2008; Levendusky 2012; Stroud 2011; Yanovitzky & Cappella 2001).

How does the availability of strident partisan media programming affect citizen attitudes and behaviors? Many observers are concerned that media fragmentation along partisan lines polarizes the citizenry and threatens the democratic system (for example: DellaVigna & Kaplan

2007; Dilliplane 2011; Holbert, Garrett, & Gleason 2010; Jamieson & Cappella 2008; Johnson &

Arceneaux 2011; Levendusky 2012; Mutz 2008; Prior 2007; Stroud 2011; Sunstein 2009; Bruce

A. Williams & Delli Carpini 2011; Yanovitzky & Cappella 2001). This fear stems from the expectation that exposure to one-sided and zealous media programming increases ideological extremism and balkanizes the population into segregated enclaves. Yet some scholars take a LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 2

more optimistic perspective on the implications of media fragmentation for democracy. They

argue that exposure to more unified and outspoken partisan views in the media may mobilize

citizens to engage in the political process, whereas exposure to more balanced or diverse perspectives may discourage participation (DellaVigna & Kaplan 2007; Dilliplane 2011;

Jamieson & Cappella 2008; Nir & Druckman 2008; Stroud 2006, 2007). In this article, we focus

on two outcomes thought to be related to consumption of homogenous and extreme partisan

programming that are representative of the negative and positive normative implications of

partisan media, respectively: (1) issue polarization and (2) participation in campaign activities.

The effect of selectively partisan media consumption can only be determined if we have a

comprehensive measure of media consumed by each individual. This article contributes to the

existing literature by developing a holistic indicator of the homogeneity and extremity of a media

diet. Existing research on partisan media tests the independent effects of exposure to a partisan

message or media segment,ii a single program or network,iii or a category of partisan media.iv

Yet we know that most Americans consume a mix of programming, with varying degrees of

bias, sometimes from opposite ends of the ideological spectrum. The net effect of media

consumption may depend on the combination of programs consumed. We use public opinion

data from the 2008 National Annenberg Election Survey (NAES) to develop a single measure of

the homogeneity and extremity of each respondent’s media consumption based on audience

composition for 73 different television and political talk radio shows included in the survey.v

Our fixed-effects and matching analyses show that more ideologically slanted media diets

are associated with greater increases in participation over the course of the campaign than

balanced media diets. People who choose strongly biased and like-minded media become more

involved in the political process than those who hear only muted calls to action from moderate LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 3

mainstream broadcasts, or those who receive mixed signals from both sides of the spectrum. We

do not find evidence that exposure to strident and unified partisan voices causes greater

polarization of attitudes on prominent issues (the Iraq War, immigration, free trade). Our results

suggest that one-sided partisan media consumption may mobilize without polarizing.

The next section of this article elaborates on the perspectives outlined above to develop

hypotheses about the effects of homogenous and extreme media diets on attitudinal polarization

and political participation. We then describe our measurement approach, data and research

strategy. We present the results from subject fixed-effects regression analyses followed by the results of matching analyses. We conclude with a discussion of key findings from the research.

Why Might Partisan Media Polarize Or Mobilize?

To develop expectations about the effects of homogeneous as opposed to heterogeneous

media diets we first review the literature on likeminded media, given that individuals with

lopsided partisan media diets most likely consume media that agrees with their political

predispositions. We develop two hypotheses based on research about the effects of likeminded

media. We then draw on studies of crosscutting media to help us theorize about the effects of

heterogeneous media diets, including both neutral (i.e. internally diverse programming) and

partisan (i.e. externally diverse programming) media from both sides. The rather defined

predictions derived from the literature on the effects of likeminded media relative to neutral

media become much murkier when we consider crosscutting media. Theoretical ambiguity about

the effects of heterogeneous media diets motivates the empirical investigation in this article.

LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 4

Effects of homogenous media diets

Theories of motivated reasoning, persuasion and social identity all suggest that exposure

to strident like-minded media will polarize citizens, especially in the absence of crosscutting

viewpoints. Partisan television and radio programs are explicitly opinionated. They frame, spin,

or slant coverage in order to support a particular perspective, highlighting favorable information

while discrediting and denigrating alternative views, organizations and personalities (Jamieson &

Cappella 2008; Levendusky 2012). Greater exposure to self-reinforcing messages can increase

confidence in one’s views, whereas decreased exposure to dissonant information reduces the

likelihood that individuals will reconsider and moderate their views (Mutz 2006). Furthermore, repeated exposure to more ideologically extreme perspectives espoused by political or media elites may persuade viewers to adopt more extreme views themselves (Feldman 2011).vi

Social identity theory suggests that media consumers may also envision themselves to be

part of a community of like-minded media personalities and audience members. Lack of

exposure to individuals with differing interests and perspectives may breed distrust and

intolerance of out-group members, causing listeners to reject attitudes associated with the out-

group in favor of those perceived to be associated with the in-group (e.g. Gutmann & Thompson

1996). This effect may be especially pronounced when the out-group is frequently denigrated

and the in-group praised (Levendusky 2012). In sum, given the nature of our changing media

environment, there seem to be sound theoretical reasons to be pessimistic about the future of our

democracy: the growth of ideologically slanted media is predicted to polarize the citizenry and

harm our ability to deliberate. To empirically evaluate this pessimistic perspective, we propose

the following hypothesis:

H1: Exposure to more homogenous and extreme partisan media diets results in greater polarization on key policy issues. LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 5

Yet as the introduction to the article indicates, pessimists are not the only voices heard in the debate about the implications of partisan news. The more optimistic theoretical perspective holds that exposure to more unified and outspoken partisan voices in the media may mobilize citizens to participate in politics, whereas exposure to more balanced media perspectives may dampen participation (DellaVigna & Kaplan 2007; Dilliplane 2011; Jamieson & Cappella 2008;

Nir & Druckman 2008; Stroud 2006, 2007). This perspective draws from theories of participatory democracy and the historical and comparative evidence of effects of party parallelism (Hallin & Mancini 2004; Schudson 1992). It also builds on more recent empirical studies of the consequences of cross-pressuring through interpersonal communication, social network location, or socio-demographic characteristics. Studies have found that homogenous social and attitudinal influences are associated with greater participation and cross-pressures associated with less participation (Brader, Tucker, & Therriault 2011; Eveland Jr & Hively 2009;

Huckfeldt, Mendez, & Osborn 2004; Huckfeldt & Sprague 1987; Lazarsfeld, Berelson, & Gaudet

1948; McClurg 2006; Mutz 2002, 2006), though others have found limited or no effects

(Leighley 1990; Nir 2005; Scheufele et al. 2006). In addition, audiences of partisan media may have more exposure to emotional appeals to get involved on behalf of their candidate or cause and information on how to get involved than audiences of neutral or mixed programming.

Moreover, the theories mentioned earlier in our discussion about polarization may also help to explain mobilization. Attitude reinforcement and polarization as a result of motivated reasoning or persuasion, are thought to foster greater activity in support of hardened and extreme attitudes (Stroud 2006, p. 151). Social identity theory suggests that individuals may be motivated to engage in activities that demonstrate their membership in a social group, whether real or imagined. Consumers of homogeneous and extreme partisan media may be motivated to LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 6 participate so as to demonstrate to themselves that they are accepted members of their (imagined) in-group of like-minded audience members and hosts.

In sum, the normatively optimistic perspective posits that increased diversification of media and partisan sorting of audiences may encourage participatory behaviors thought to be beneficial for democracy. To empirically evaluate the more optimistic perspective, we forward an additional hypothesis:

H2: Exposure to more homogenous and extreme partisan media diets results in higher levels of political participation.

So far, empirical testing of the effect of likeminded partisan media is fairly limited, but existing evidence provides some support for the view that it increases political polarization

(DellaVigna & Kaplan 2007; Groseclose & Milyo 2005; Holbert et al. 2010; Jamieson &

Cappella 2008; Levendusky 2012; Mutz 2008; Stroud 2006, 2010) and participation (DellaVigna

& Kaplan 2007; Dilliplane 2011; Jamieson & Cappella 2008; Nir & Druckman 2008; Stroud

2006, 2007). But while many experimental and survey studies show partisan media effects, other scholars have found very small or no effects (Gerber, Karlan, & Bergan 2009; Meffert et al.

2006).

A variety of comparisons are invoked within the theoretical and empirical research on media effects. Likeminded media are often compared to neutral media effects. Sometimes attitude congruent stimuli are compared to the absence of stimuli, or to similar media without partisan cues. Occasionally the effects of likeminded media are compared to those from crosscutting media. To understand the effect of homogeneous as compared to heterogeneous media diets, we also need to consider more explicitly the nature and effects of consuming media that includes varying degrees of conflicting arguments.

LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 7

Effects of heterogeneous media diets

Theoretical predictions about the effects of homogeneous partisan media consumption

(i.e. consumption of only likeminded media) are well developed in the literature and relatively straightforward. In contrast, the outcomes resulting from heterogeneous media diets are largely untheorized and untested. To deduce the likely outcomes of consuming a mixture of likeminded, crosscutting and neutral programming we draw on studies of crosscutting media effects. In doing so, we consider how the consumption of crosscutting media in combination with likeminded media might produce an effect that is more than just the sum of its parts.

Research on information processing suggests that individuals are motivated by two goals: accuracy, which compels a search for correct judgments, and the preservation of preexisting beliefs, which compels a search for preferred results (Taber & Lodge 2006). Depending on which motivation prevails, and how strongly, three different scenarios are plausible for how individuals will react to heterogeneous as opposed to homogeneous media diets.

First, if individuals are motivated to seek accurate information, then exposure to diverse perspectives in the media will cause listeners to reconsider and moderate their initial views or behaviors. Under this scenario, we would expect individuals with more heterogeneous media diets to be less extreme and active than those with homogeneous media diets. This first scenario is consistent with our hypotheses described above.

Second, if individuals pursue partisan directional goals, then they will process incoming political information in ways that favor their initial partisan predispositions. Disconfirmation bias may cause individuals to dismiss, or even ignore, alternative perspectives. Theories of persuasion suggest that people are often immune to influence from crosscutting media since they are less likely to see such media as credible and trustworthy (Levendusky 2012). Accordingly, LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 8

we might expect that individuals who consume media from both sides of the spectrum would be

identical to individuals who consume from one side of the spectrum; the pro-attitudinal messages

polarize while the counter-attitudinal ones have no effect.

Finally, individuals with especially strong partisan information processing goals may

seek out crosscutting information to serve as a foil for their preferred arguments. Exposure to

inconsistent views can prompt counterargument, which in turn strengthens initial attitudes (Taber

& Lodge 2006). Social identity theory suggests a similar polarizing response when individuals

are exposed to criticisms of their own partisan in-group from media affiliated with the other side

(Levendusky 2012). In this final scenario, that partisan motivated individuals who consume media from both sides of the spectrum will become more polarized and participatory than

individuals who consume from only one side of the spectrum.

The three scenarios described here suggest three different hypotheses about the difference

between homogeneous and heterogeneous partisan media diets. While all three alternatives are

plausible, we believe the first to be most likely for our current investigation. Unlike experiments

in which some respondents are forced to watch counter-attitudinal programs that they dislike or

distrust, the respondents in our observational data choose exposure to opposing perspectives (or

at least they do not avoid it) (Arceneaux and Johnson 2008). We think these respondents are

more likely to watch diverse media because they have a positive or neutral view of the shows

(despite seemingly conflicting messages) than as a deliberate attempt to counter-argue or ignore the presented material (Levendusky 2012, p. 21). For this reason, we expect that the depolarizing and demobilizing effect of exposure to different perspectives for these media consumers will swamp the effects of those who are motivated to counter-argue or those who are inoculated by distrust from the effects of counter-attitudinal media in a diverse media diet.vii LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 9

Therefore, we posit that the effects of an externally diverse media diet (i.e. consumption of

extreme media from both sides of the spectrum) will have similar depolarizing and demobilizing

effects as an internally diverse media diet obtained through the consumption of neutral media. In

short, we expect that a homogenous and extreme media diet will be more polarizing and

mobilizing when compared to either a more neutral diet or a more diverse diet. However, we

note that there are multiple theoretically plausible outcomes and we thus turn to the empirical

investigation to help adjudicate between the different scenarios.

Before moving on, it is important to note that the pessimistic and optimistic perspectives

regarding the effects of media fragmentation and the re-emergence of partisan news are not

mutually exclusive, as partisan media can both polarize and mobilize American citizens. In fact,

as Mutz (2006) notes, the same psychological mechanisms may be at work in both processes.

Assuming that our hypothesized effects are supported by empirical evidence, the normative implications of partisan media may depend on whether one prioritizes a deliberative or participatory theory of democracy. However, if the different hypotheses are not equally supported by the data then the suggested tradeoff between harmony and engagement may not be necessary. With this in mind we now turn to the empirical evidence to evaluate the proposed hypotheses.

Measures and Methods

This study uses data from the Internet Panel of the 2008 National Annenberg Election

Survey (NAES). The NAES panel included five waves of data collected from a nationally representative sample of American adults. The first wave was conducted before the primaries and the last took place soon after the election. We were able to draw on a sample of 4,052 LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 10

respondents who answered the questions used to create our independent variables.

Independent Variable: Media Partisanship Score

Testing our hypotheses requires a new measure of exposure that incorporates a variety of

politically relevant television and radio programs. Our measure is constructed from answers to

survey questions asking about exposure to a total of 73 programs.viii Participants were shown lists of approximately 15 shows at a time, and asked to indicate shows that they watched or listened to “regularly” or “at least once a month”.ix

We employ these questions to develop the Media Partisanship Score (MPS), a novel

measure indicating the homogeneity or heterogeneity of each respondent’s media diet. In brief,

we first estimate a continuous indicator of partisan bias for each television or radio program

based on audience preferences.x We then use this indicator of program bias to predict the

partisan bias of each respondent’s reported media diet.

Specifically, we developed our measure from a method for measuring sociodemographic

cross-pressures developed (and validated) by Brader, Tucker, and Therriault (2010; 2011),

adapting their methodology to measure media consumption rather than sociodemographic

characteristics.xi MPS is constructed using a four step process.

1. In the first step, we used our 73 measures of media exposure from Wave 2 to predict

preference for McCainxii during Wave 2 using logistic regression. We see this first step as

generating coefficient estimates that describe the direction and intensity of each media

program’s partisanship based on the partisan preferences of their audiences. For

example, the coefficients for the shows hosted by Rush Limbaugh or Keith Olbermann LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 11

were large in magnitude, but opposite in direction. The estimated coefficients for each

show are included in the Supporting Information.

2. In the second step, we used the coefficient estimates generated in Step 1 to calculate the

predicted probability for each respondent to prefer McCain (based on media

consumption) while suppressing the constant term generated in Step 1.xiii

3. In the third step, we calculated the predicted probability of preferring Obama as 1 minus

the probability of preferring McCain. For example, a person whose predicted probability

of voting for McCain during the primaries was .9 (Step 2) would have a predicted

probability of voting for Obama of .1 (Step 3).

4. In the fourth step, we calculated the absolute value of the difference between these two

predicted probabilities. To use the example above, the difference between these

probabilities would be .8.

The four-step process results in the MPS, a measure bounded between 0 and 1.

Individuals with an MPS close to 1 were thus individuals whose media diet strongly favors one

candidate, indicating that they consumed media that was strongly partisan from one side of the

ideological spectrum without any counter-attitudinal consumption. In contrast, people with values close to 0, had media diets that were balanced, consuming media that was neutral and/or equally represented both sides of the partisan spectrum. Given how we have constructed the measure, a respondent might have a low MPS because they watch internally diverse shows LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 12

(neutral media which presents both sides of the ideological spectrum on the same program) or

externally diverse shows (partisan media selected from both sides). A low MPS thus indicates

that the individual is exposed to a cross-pressured media diet.

We repeated the four-step process initially used to generate MPS for Wave 2 to generated a new MPS during Wave 4. We make one change in calculating the MPS for Wave 4. In the first step, instead of generating a new set of coefficient estimates for each program using only Wave 4 data, we used the same coefficient estimates for each program that were previously generated using Wave 2 data. For example, the predicted effect of watching Program X might be b=.800 according to the coefficients generated with data from the wave 2. In step 2, b=.800, which would then be multiplied by watching (1)/not watching (0) a program in wave 4. The only thing that changes in our calculation of the MPS between Wave 2 and Wave 4 is whether or not a

person watches Program X (i.e., whether it is coded as “1” or “0”) in the relevant wave.

Therefore, change in a respondent’s MPS between the waves results solely from their changing

media habits over time rather than a change in the estimated indicators of the bias in each show.

Advantages of the Media Partisanship Score

Our MPS measure offers five innovations. First it allows us to evaluate the interactive

effects of different programs by recording the homogeneity or heterogeneity of a consumer’s

total reported media diet. Most research tests the independent effects of each media type in

isolation. Second, it is based on a continuous scale indicating the degree of each program’s

partisan bias (from extreme left to extreme right). Other measures rely on a discrete

categorization each program’s type (Democratic/Republican, Liberal/Conservative). Third, MPS

is equally valid for independents and partisans. Existing measures of likeminded and crosscutting LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 13

media must exclude independents because they cannot be coded to fit the categories.xiv

Fourth, our measure is transparent and replicable across time or media and political

systems. Alternative measures rely on audience perception or secondary data sources to code the

partisanship of media programs. Secondary sources on program bias are often incomplete or

unavailable for a given time period or country. Choosing sources and strategies for filling in

gaps requires judgment calls by the analyst. As partisan bias often lies in the eye of the beholder

(as argued by the hostile media thesis), coder attitudes can shape the resulting coding of partisan

bias (Vallone, Ross, and Lepper 1985). Respondent perceptions are also subject to individual

biases.

Fifth, MPS incorporates the influence of 73 TV and radio shows spanning entertainment

and news genres into a single metric. Other analyses are typically limited to hard news programs

where evidence on bias is obtainable. Our methods can include a large selection of news,

entertainment (e.g. 24 or The Simpsons), satirical news (e.g. The Daily Show) and soft news programs (e.g. Good Morning America).

MPS is a valuable addition to scholarship on partisan media for the five reasons outlined

above. Ultimately, we conceive of our measure of media diets as a compliment to (rather than a

replacement for) extant measures of individual programming types, given that we seek to

measure a difference concept than previous scholarship.

MPS Measure Validity

For the MPS to be a useful representation of an individual’s media consumption choices,

we must establish that the MPS serves as a valid measure of the homogeneity or heterogeneity of

a person’s media diet. As Brader, Tucker, and Therriault (2011) note, it is impossible to directly LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 14 measure an individual’s experience of cross-pressures, so we lack an absolute standard against which to compare our method. However, establishing that MPS scores are not subject to significant variation based on researcher assumptions or the size of our dataset, comparing it to available existing measures, and establishing that it conforms to basic expectations can increase our confidence in its validity. These validity checks also help to establish the appropriate bounds for use of the method.

To establish the robustness of the MPS against researcher choices, we examined the effect of changing the battery of media programs from which the measure was generated. This test is prompted by two concerns: (1) that inclusion of outlier media programs consumed by few respondents may exhibit undue influence on the calculation of the MPS and (2) that the media programs included in future surveys may differ from those included in the NAES in a manner that affects the calculation of the MPS score.

To examine the sensitivity of the MPS measure to program inclusion, we first ranked the

73 programs included in the NAES from “most watched” to “least watched” based on responses.

We then recalculated the MPS for wave 2 (using the process described above) with subsets of the most popular media programs, testing the effect of including between 5 and 50 media programs

(in increments of five) in the calculation of our measure. After calculating this restricted MPS, we correlated the result with the unrestricted MPS, calculated using all 73 media programs.

Figure 1 below shows the results of this test, which indicate that including just over half of the available media programs (n = 40) results in a strong correlation (ρ = 0.90) with the unrestricted

MPS score. At the same time, the sizeable increase in correlation between the score calculated using 35 media programs and that calculated using 40 suggests that researchers should be wary of calculating MPS scores based on datasets that do not include an extensive battery of media LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 15 programs from which the respondents can choose. LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 16

FIGURE 1. Correlation between restricted MPS scores calculated using a subset of media programs and the unrestricted MPS score using 73 programs

LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 17

We also examined the sensitivity of the MPS to changes in the dataset, to determine

whether the scores calculated using a subset of respondents would vary significantly from those

using the full pool of respondents available in our data. To do this, we randomly sampled subsets

from the data (varying from 1000 to 5000 respondents, in increments of 1000), calculated a

restricted MPS based on this subset, and calculated the correlation between the restricted MPS

and the MPS based on the full dataset. We repeated this 1000 times, allowing the calculation of the mean correlation and a 95% confidence interval. As shown below in Figure 2, MPS scores calculated based on subsets as small as 1000 respondents result in highly correlated MPS scores

(ρ = 0.83 ± 0.05), increasing above 0.9 for sample sizes of 2000 respondents or more. Similarly high correlations (ρ = 0.94 ± 0.02) were found when the dataset was split, with MPS scores for

one half of the respondents calculated based on the media consumption and candidate

preferences of the remaining respondents. The robustness with respect to changes in media

programs and sample is an encouraging sign for the validity of the MPS. LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 18

FIGURE 2. Correlation between restricted MPS scores calculated using a sample of respondents and the unrestricted MPS score using all respondents

LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 19

Of course, robust scores that do not accurately reflect some basic expectations about

respondents who watch different media programs would be useless. As a baseline check, we

compared the average MPS scores among populations in which we would expect to see

significant differences, examining whether the differences were evident in the data and occurred

in the expected direction. The first test compared the average MPS score among respondents

who indicated that they consumed both “Countdown with Keith Olbermann” (MSNBC) and

“The O’Reilly Report” (FNC) against those who indicated they watched either “Countdown” or

“The O’Reilly Report” but not the other. Given the manner in which our measure is constructed,

we would expect the former group to have a lower MPS (i.e. consume a less partisan news diet)

than the latter. This expectation is confirmed, as respondents who indicated that they watched

both shows had an average MPS of 0.49, significantly less than the average (0.72) among those

who watched either show (p<.001). Similarly, a comparison of respondents who listened to both

NPR’s All Things Considered and The Rush Limbaugh Show to those who watched one but not

the other showed comparable results, with average MPS scores of 0.54 and 0.64, respectively (p

< .001).

Finally, we examined the extent to which our estimated coefficients for the bias of each

show based on respondent revealed preferences are similar to other measures of program bias

based on respondent perceived bias. Dilliplane (2011) divided media programs into programs

with a Republican slant, programs with a Democratic slant, and neutral programs based on a

separate 2008 NAES cross-sectional survey that asked respondents which candidate was favored by the program from which they got most of their news about the 2008 presidential election. If more than 25% of regular watchers identified the show as having a particular slant, it was classified accordingly. Less than 25% of regular watchers indicating that the program was geared LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 20

toward either party resulted in a classification of “neutral program.” Figure 3 plots our estimates

of program bias (the estimated log-odds coefficients) for a selection of media programs along a

continuum from -1 (partisan towards Obama) to +1 (partisan towards McCain). Dilliplane’s

(2011) codings of bias are indicated by the shading of the media program label.xv Generally, the

shows align as we would expect, with Republican-slanting shows on the right and Democratic-

slanting shows on the left. LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 21

FIGURE 3. Comparison of log-odds coefficients for media programs predicting a preference for McCain and partisan slant from Dilliplane (2011)

(Note: Programs identified as having a Republican slant are labeled in black, Democratic slant in white text on a black background, and neutral in grey.)

LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 22

In sum, the full set of validity checks shows that the MPS (1) is not very sensitive to the choice of media programs used in its calculation (above a certain baseline); (2) does not vary significantly when calculated using population subsets; (3) confirms certain basic expectations about direction of differences among sample populations; and (4) with a few exceptions, calculates similar partisan slants as those indicated by survey respondents. These characteristics of the MPS suggest that it can serve as a valid measure of the partisanship of someone’s media diet. The predictive validity of our measure is demonstrated with analyses of the effects of MPS on participation described below.

Dependent Variables: Issue Polarization and Campaign Participation

Our measure of issue polarization was generated from four issue questions (relevant in the 2008 election) which prompted respondents to give their views on the war in Iraq, a path to citizenship for illegal aliens, the security of border with Mexico, and free trade.xvi For each question, respondents were presented with answers that spanned the conventional views on the issues, and were considered “polarized” on an issue if they chose an extreme response on either side of the spectrum. On average, respondents held more polarized opinions during the convention season (M=1.675; SD=1.118) than they did during the general election (M=1.626,

SD=1.109), which means that respondents became less extreme over time.

Campaign participation was measured using five items traditionally used to measure participation in elections. Respondents reported whether, in the past 12 months, they had (1) given money to a campaign; (2) worked or volunteered for a campaign; (3) persuaded others in favor of a candidate; (4) attended a political meeting, speech, or event; or (5) advertised their support for a candidate via buttons, stickers, or signs. The individual items were summed into a LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 23

5-point index and analyzed using linear models. On average, respondents reported that they had

participated in one campaign activity during the convention season (M=.998; SD=1.214), which

increased to an average of between one and two activities following the general election

(M=1.445; SD=1.440).

Research Design

Subject Fixed-Effects Design

First, we used panel survey data to estimate subject fixed-effects models. Due to the difficulty of correctly implementing fixed-effects models using count data (Allison, 2009), we employed a linear model in which

yit = β0 + β1MPS it + β2Total # of News Programs it +αi + uit (1)

where y is the outcome of interest, MPS it provides respondent i’s media diet at Wave t, Total # of News Programsit provides the total number of news programs respondent i watched at Wave

t,xvii α is subject fixed effects, and u is an unobserved disturbance term. The benefit of these

fixed-effects models is that they control for unobserved heterogeneity, thus ruling out concerns related to spuriousness (if the unobserved characteristics are time invariant). This is to say that some subjects may be more politically interested, more knowledgeable, etc. than others—but, by

virtue of fixed-effects models, we are controlling for these stable characteristics whether they are

observed or unobserved. In essence, we are comparing subjects with themselves at an earlier

time. Using this approach, β1 provides the within-subjects effect of the media partisanship score,

indicating the effect of changes in media diet over time within individuals.

We face a common challenge of using observational data for causal inference, that of

being unable to establish the direction of influence with certainty. Nonetheless, we designed our LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 24

analysis such that the dependent variable was measured after the initial measure of the

independent variable. Specifically, issue polarization was measured during Wave 3 (the

convention season) and Wave 4 (the general election). Campaign participation was measured

during Wave 3 (the convention season) and Wave 5 (the post-election period). The independent variable MPS, was measured during Wave 2 (the primaries) and Wave 4 (the general election).

This lag design can help somewhat with establishing causal order. In addition, the effects of media often occur sometime after consumption, rather than contemporaneously, and the lag better accounts for possibility of delayed effects.

Matching Design

As with any technique, our regression models are not perfect. To examine the robustness

of our findings, we also examine the effects of changes in MPS over the course of the survey

when matching subjects on a set of covariates. Matching analyses reduce concerns that

determinants of the subject’s media diet (and thus of their MPS) also influence their political

participation and issue polarization because we only compare subjects that are similar on a large

set of observables (Dehejia & Wahba 1999; Barabas 2004; Kam & Palmer 2008; Ladd & Lenz

2009; Meredith & Malhotra 2011; Levendusky 2011). “Treatment” and “control” groups were

matched on a set of demographic characteristics, MPS in Wave 2, and politically relevant

variables. A description of the variables on which respondents were matched is included in the

Supporting Information.

Matching requires the definition of a binary “treatment” to separate the two groups. We

perform three separate matching analyses, dividing respondents into “High MPS” and “Low

MPS” groups. For the three analyses, “High MPS” groups are defined as subjects having MPS in LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 25

Wave 5 that are: (1) above the lowest tercile; (2) above the median; and (3) in the top tercile

(Levendusky, 2011). Since respondents are matched based on their MPS at Wave 2, the

matching analysis allows the comparison of two demographically similar respondents whose

MPS measures diverged during the course of the election.

The motivation behind matching analysis is to generate “control” and “treatment” groups

that are similar in all respects but their treatment status by matching each treated observation to a

corresponding control observation. Tables SI.6-SI.8 in the online appendix compare differences

in mean values between the two groups in the unmatched data, under unrestricted Mahalanobis

matching, and under Mahalanobis matching with a caliper restrictionxviii. Differences in the MPS at Wave 2 are significantly less in the matched data than they are in the unmatched data.

Mahalanobis matching with a caliper restriction is able to reduce group-wide differences in means of matched variables in the control and treatment groups to the point where none are significantly different at the 5% level.

Once the matched groups are generated, we perform a simple comparison of means to determine whether being in the “treatment” group (top two terciles, above the median, or top tercile of MPS at wave 5) is associated with an increase in political participation and ideology relative to those in the “control group” (bottom tercile, below the median, or lower two terciles of MPS at wave 5).

Results

Subject Fixed-Effects Regression Results

Table 1 below reports the coefficient estimates generated from Equation 1, which

presents the fixed-effects models that control for between-subject differences in media diet, as LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 26 they pertain to our first hypothesis regarding issue polarization and campaign participation. We initially hypothesized that an increasingly homogenous and extreme media diet will increase issue polarization (H1). The results in Table 1 indicate that our hypothesis about the relationship between media diet and issue polarization is not supported by the data (b=-.000, SE=.066). The fixed-effects regression indicates no correlation between within-subject change in MPS and ideological extremity across our two waves. It seems that homogenous and extreme media diets do not contribute to issue polarization.

LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 27

TABLE 1. Effects of MPS on Issue Polarization and Campaign Participation

Polarization Participation Within-subjects MPS -.0000 .220*0 (.066) (.069) Total media .000 -.0030 (.004) (.004) Constant 1.645*) 1.138*** (.054) (.050) Observations 6866 8104 Groups 3433 4052 R2 overall .008 .070 * p<.05

LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 28

The results in Table 1 do support our argument about media diet and campaign

participation. We postulated that homogenous and ideologically extreme media diets result in

increased campaign participation (H2). The fixed-effects models show that, on average, a person who moved from a more cross-pressured or neutral media diet to a more homogenous and extreme media diet would participate more in the election campaign (b=.220, SE=.069). During the general election, the size of this effect indicates that respondents with MPS (M = 0.557, SD =

0.298) of one standard deviation above the mean will participate in 0.113 additional campaign activities on average.

Matching Analyses Results

Full results of the matching analyses are shown in Table 2 for all three of the treatment

definitions (lower tercile, above median, and upper tercile) considered across a range of

matching criteria. Tables 2 and 3 show the comparison of means with respect to issue

polarization and campaign participation without matching, matching without a caliper restriction,

and matching with a caliper restriction of 0.25 standard deviations of the Mahalanobis distance. LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 29

TABLE 2 Comparison of covariate means for unmatched, matched without caliper, and matched with caliper data sets

Issue Polarization Mean for: N = Treatment Control Difference Treated Control Treatment 1 (above lower tercile) .0740 2,036 970 Unmatched 1.679 1.605 (.043) -.015 2,036 970 Matched, No Caliper 1.679 1.694 (.073) .039 230 970 Matched, Caliper 1.713 1.674 (.127)

Treatment 2 (above median) .134* 1,526 1,480 Unmatched 1.721 1.587 (.041) .049 1,526 1,480 Matched, No Caliper 1.721 1.672 (.071) .174 276 1,480 Matched, Caliper 1.913 1.739 (.114)

Treatment 3 (in upper tercile) .133*0 1,043 1,963 Unmatched 1.742 1.609 (.043) .035 1,043 1,963 Matched, No Caliper 1.742 1.708 (.071) .047 295 1,963 Matched, Caliper 1.946 1.898 (.111)

Note: * p<.05, two-sided T-test with N-2 df. Standard errors are shown in parentheses below the differences

LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 30

TABLE 3. Comparison of covariate means for unmatched, matched without caliper, and matched with caliper data sets

Campaign Participation Mean for: N = Treatment Control Difference Treated Control Treatment 1 (above lower tercile) .752* 2,167 1,030 Unmatched 1.814 1.062 (.053) .376* 2,167 1,030 Matched, No Caliper 1.814 1.438 (.086) .322* 261 1,030 Matched, Caliper 1.704 1.383 (.151)

Treatment 2 (above median) .832* 1,633 1,564 Unmatched 1.979 1.147 (.049) .374* 1,633 1,564 Matched, No Caliper 1.979 1.605 (.089) .396* 308 1,564 Matched, Caliper 1.990 1.594 (.145)

Treatment 3 (in upper tercile) .880* 1,113 2,084 Unmatched 2.145 1.266 (.052) .288* 1,113 2,084 Matched, No Caliper 2.145 1.858 (.090) .067 314 2,084 Matched, Caliper 2.217 2.150 (.149)

Note: * p<.05, two-sided T-test with N-2 df. Standard errors are shown in parentheses below the differences

LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 31

As expected, there are differences in the means between the different unmatched groups indicating that individuals who choose to watch homogenous and extreme media on average have different opinions than those with more diverse or neutral media diets. However, the differences reflected in these unmatched comparisons do not account for confounding factors.

Table 2 shows that when individuals are matched, there is no significant difference in the means for issue polarization between those with high and low MPS. The results are insignificant regardless of whether the matching is done without a caliper or with the caliper, and regardless of how the “treatment” is defined. This is consistent with the findings of the regression analysis, which indicated that there was no association between MPS and issue polarization in our data.

Once again, there is no evidence to support the first hypothesis that homogenous and extreme media diets increase issue polarization (H1).

Similarly, Table 2 supports the findings of the regression analysis above with regards to political participation in two of the three treatment definitions. Over the course of the campaign, increasing MPS from the lowest tercile to the upper terciles or from below the median to above the median results in increased political participation regardless of whether a caliper is used. For treatment 3, the effect is significant only for the analysis without the caliper, but not with the caliper. Overall, the results of the matching analyses provide additional support for our hypothesis that a homogenous and extreme media diet energizes political participation (H2).

We used multiple methods to account for threats to causal inference; however, our study is not without limitations. We cannot fully rule out the possibility of reverse causation without experimental data that would ensure temporal precedence. Moreover, we chose analytical methods that minimize the likelihood of spuriousness—namely, by using fixed effects and matching—however, there are still potential threats to causal attribution. Specifically, our efforts LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 32

to match on key variables are only as good as those variables that we have observed and all

fixed-effects models are vulnerable to the criticism that they do not control for spuriousness if

the unobserved variables are time-varying instead of time-invariant. Nonetheless, our claim that campaign participation is boosted by an increasingly homogenous and extreme media diet is strengthened by the consistent evidence generated from a variety of analytical methods.

Discussion

This article provides a methodological contribution with the introduction of a new measure, and an empirical contribution by analyzing the effects of one-sided versus balanced or neutral media diets. The measure we developed, MPS, offers five benefits. First, it provides a holistic measure of the homogeneity of media diets that facilitates analysis of the interactive effects of numerous media programs. Second, it avoids classifying programs simply by partisanship, and instead reflects the degree of each program’s partisanship in a continuous measure ranging from extreme left to extreme right. Third, it applies to independents as well as to partisans (who themselves may vary in strength of partisanship). Fourth, it can be replicated overtime and in international contexts. Fifth, our method is well suited to the study of media in the new information environment. It can accommodate numerous and diverse types of politically relevant information sources, including entertainment, soft news and satirical programs along with hard news (Dilliplane, Goldman, and Mutz 2013). Our method can also integrate media consumed on television, radio, or the internet into a single measure.

We use our new measure to evaluate the hypothesized beneficial and harmful political effects of unbalanced versus balanced media diets. Our subject fixed-effects regressions and matching analyses suggest that consuming an ideologically homogenous and extreme media diet LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 33 increases campaign participation. We find no empirical support for the claim that a lopsided partisan media diet polarizes citizen opinions on key issues. As such, our findings lend credence to those optimists who hope that the prevalence of partisan media choices will boost political participation, while calling into question the concerns of pessimists who predict that niche programs will lead to ideological polarization. The combination of results suggest that unbalanced media diets increase participation because individuals are pulled into the political process by partisan appeals to get involved, and targeted information about how to do so.

Concerns that individuals are mobilized only by increased opinion strength may be unwarranted.

Our study has several theoretical implications. Research on exposure to like-minded media seems insufficient for understanding the effects of homogeneous as compared to heterogeneous media diets on opinion polarization. Fortunately, cross-cutting media effects research offers several plausible explanations for the null results with respect to the polarizing effect of slanted media diets. First, it may be that individuals simply ignore perspectives that contradict their initial views. If this is the case, then individuals who consume media that presents both attitude-consistent and attitude-inconsistent arguments would have similar attitudes to those that are exposed to only agreeable perspectives. Second, when faced with diverse perspectives, some individuals may polarize as expected, while others may moderate their issue stances. Likeminded media may motivate some individuals to counterargue against cross-cutting media and provide them with fodder with which to do so. In contrast, cross-cutting media may persuade other individuals to reconsider and temper initial opinions. The estimated average effect of positive reactions by some individuals, and negative reactions by others, would be consistent with our null result. Additional research is necessary to determine which of these explanations are most plausible.xix LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 34

Overall, our study may help assuage the tension between the competing theories of deliberative and participatory democracy in the new media environment. Our findings should give hope to advocates of participatory democracy, as increasing media diversification should lead to increased political participation, thus opening the door to further beneficial outcomes stemming from citizen engagement. At the same time, our findings also indicate that people do not become more polarized in their issue positions as a result of consuming unbalanced partisan media diets. Further work on the effects of partisan media is merited to refine our understanding of how these competing dynamics play out in today’s increasingly partisan news landscape.

LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 35

Endnotes

i For example, 35 percent of Democrats reported that they watched FOX News in 2010

(Levendusky 2012, p. 10). Garrett (2009) argues that people seek out congenial perspectives but do not actively avoid antagonistic perspectives. Therefore, it is not surprising that more than a third of Democrats report watching Fox news. However, the 2010 figure suggests a recent decline in Democratic viewership of Fox. Data from 2000 to 2008 reveals that 45 percent or more of Democrats reported watching Fox (Levendusky 2012, p. 18).

ii For example, lab experiments that test the effects of a single news segment.

iii For example, Barker (2002) examines the relationship between statements made by Rush

Limbaugh and listeners’ opinions, activities, and beliefs.

iv For example, Dilliplane (2011) measures the conditional effects of like-minded or cross-cutting media exposure.

v To create this measure we build off of a method developed by Brader, Tucker, and Therriualt

(2010, 2011) to measures socio-demographic cross-pressures.

vi For empirical demonstrations, see Abelson (1995) and Downing, Judd, & Brauer (1992).

vii Relatedly, some have argued that experimental studies systematically overestimate the actual

attitudinal and behavioral effects of partisan media because they do not take into account the fact

that many people avoid news altogether, or react differently to media when allowed to choose

programs (Arceneaux & Johnson 2008; Johnson & Arceneaux 2011; Prior 2007).

viii Forty of these shows were classified as “news,” with 24 appearing on cable stations (e.g.

FNC, MSNBC, and CNN) and 16 appearing on network news (ABC, NBC, CBS, and PBS).

Eighteen programs were classified as non-news content, and ranged from dramas such as 24 and LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 36

CSI: Miami to talk shows such as Oprah and The Ellen Generes Show. Finally, questions

regarding 15 political talk radio shows were asked of a random subsample of respondents

(n=5251). The vast majority of these programs were hosted by prominent conservatives (e.g.

Glenn Beck), with the exception of NPR’s “All Things Considered.” Programs that were

included on the survey were selected due to their popularity according to Nielson ratings prior to

the election.

ix Respondents could check off any number of programs that they wanted, with the average respondent reporting that they consumed between 10 and 11 programs during both the primaries and the general election (M=1.608; SD=7.064 during Wave 2; M=1.436; SD=6.819 during Wave

4). For a complete list of programs included, see SI.1.

x Using customer preferences to assess the bias assumes that broadcasters respond to demand- side cues from consumers when developing content (Mullainathan and Shleifer 2005; Gentzkow and Shapiro 2006).

xi The methodology developed by Brader, Tucker, and Therriault (2010, 2011) is as follows.

First, they predict the probability of voting for the two main party candidates in U.S. presidential elections using demographic and attitudinal measures. Then, they calculate the difference between those two estimated probabilities. Finally, they subtract this difference between the two predicted probabilities from 1 to rescale the measure so that higher values represent “cross-

pressures” rather than certainty. By this logic, an individual who is squarely in one camp (based

on his sociodemographic profile) is not cross-pressured and has a value close to 0, whereas an individual who is torn between the two candidates (again, based on her sociodemographic profile) is highly cross-pressured and has a value close to 1. When constructing MPS, we do not LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 37

include the final inversion step. So individuals who consume unified partisan media have values

close to 1, and individuals who consume cross-pressured media are close to 0.

xii Preference for candidates was coded following Brader, Tucker, and Therriault (2010, 2011).

First, we used vote intention to identify McCain and Obama supporters. If a respondent reported

no preference or that they did not know, we coded their preference according to party

membership (e.g. a strong Democrat would be coded as favoring Obama) or through the party they leaned towards (e.g. a person who leaned Republican would be coded as favoring McCain).

For a complete justification of this methodology, see Brader, Tucker, and Therriault (2010,

Appendix V, p. 10-12).

xiii We suppressed the constant term because we are interested in calculating the predicted bias of

a given media diet, and not the bias of a given respondent. In more technical terms, our reasons for including the constant in Step 1 but then suppressing it in Step 2 are as follows: suppressing the constant in Step 2 prevents bias in our initial estimates for the effects of media due to general preference for Obama in the dataset, but including the constant in Step 1 avoids forcing our media variables to soak up the variation caused by this preference. In suppressing the constant term in Step 2, we deviate from the methodology used by Brader, Tucker, and Therriault (see

2010, Appendix V, p. 10-12).

xiv Studies of the effects of Left, Right and Neutral media effects as opposed to likeminded and

crosscutting media effects can also include independents.

xv Note: Dilliplane (2011) does not characterize NPR’s All Things Considered or The Rush

Limbaugh Show, but those programs are included for reference.

xvi The exact question wording can be found in the Supporting Information. LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 38

xvii We controlled for the total number of shows watched to account for changes in the volume of a person’s news consumption as well as the overall tenor of that media diet. xviii Using a caliper drops treatment observations from the dataset if they are unable to be closely matched with a control observation. This reduces the sample size but increases the comparability of matched observations, strengthening the assumption of “as-if” random assignment. xix As always, measurement bias or noise may account for null effects. It may be that individuals who select unified programing are already extremely polarized for other reasons and thus do not become even more polarized in response to media. However, a ceiling effect seems unlikely given our evidence. While those respondents in the top 1/3 of MPS scores show slightly more polarized attitudes, only 7.2% held extreme positions on all four issues, leaving 92.8% of respondents with the ability to increase their polarization. Finally, it may be that issue positions are fairly stable and media effects of any kind are too small to detect with our analysis. References

Abelson, Robert P. 1995. "Attitude extremity". Attitude strength: Antecedents and

consequences, Ohio State University series on attitudes and persuasion, Vol. 4. (pp. 25–

41). Hillsdale, NJ, England: Lawrence Erlbaum Associates, Inc.

Arceneaux, Kevin, and Martin Johnson. 2008. "Choice, Attention, and Reception in Political

Communication Research." Presented at the annual meeting of the International Society

for Political Psychology, Paris, France.

Barabas, Jason. 2004. “How Deliberation Affects Policy Opinions.” American Political Science

Review 98 (04): 687-701. doi:10.1017/S0003055404041425.

Barker, David C. 2002. Rushed to Judgment: Talk Radio, Persuasion, and American Political

Behavior. New York: Columbia University Press.

Baum, Matthew A. 2002. "Sex, Lies, and War: How Soft News Brings Foreign Policy to the

Inattentive Public." American Political Science Review 96 (01): 91–109.

doi:10.1017/S0003055402004252

Baum, Matthew A. 2004. "Circling the Wagons: Soft News and Isolationism in American Public

Opinion." International Studies Quarterly 48 (2): 313–338.

Baum, Matthew A., and Angela S. Jamison. 2006. "The Oprah Effect: How Soft News Helps

Inattentive Citizens Vote Consistently." The Journal of Politics 68 (04): 946–959.

doi:10.1111/j.1468-2508.2006.00482.x

Brader, Ted, Joshua A. Tucker, and Andrew Therriault. 2010. The Cross-Pressured Citizen.

Brader, Ted, Joshua A. Tucker, and Andrew Therriault. 2011. "Cross-Pressures and Political

Participation." Presented at the Midwest Political Science Association, Chicago, IL.

Dehejia, Rajeev H., and Sadek Wahba. 1999. “Causal Effects in Nonexperimental Studies: LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 40

Reevaluating the Evaluation of Training Programs.” Journal of the American Statistical

Association 94 (448): 1053–1062.

DellaVigna, Stefano, & Ethan Kaplan. 2007. "The Fox News Effect: Media Bias and Voting."

The Quarterly Journal of Economics, 122 (3): 1187–1234.

Delli Carpini, Michael X. 2011. "Entertainment and Political Learning." In The SAGE Handbook

of Political Communication, eds. Holli A. Semetko & Margaret Scammell. Thousand

Oaks, CA: SAGE Publications Ltd., p. 650.

Delli Carpini, Michael X., & Bruce A. Williams. 2001. "Let us infotain you: politics in the new

media environment." In Mediated politics: communication in the future of democracy,

eds. W. Lance Bennett and Robert M Entman. Cambridge University Press, 160–181.

Dilliplane, Susanna. 2011. "All the News You Want to Hear: The Impact of Partisan News

Exposure on Political Participation." Public Opinion Quarterly 75 (2): 287-316.

Dilliplane, Susanna, Seth K Goldman, and Diana C. Mutz. 2013. "Televised Exposure to

Politics: New Measures for a Fragmented Media Environment." American Journal of

Political Science. 57: 236-248. doi: 10.1111/j.1540-5907.2012.00600.x

Downing, James W., Charles M. Judd, and Markus Brauer. 1992. "Effects of repeated

expressions on attitude extremity." Journal of Personality and Social Psychology 63 (1):

17–29. doi:10.1037/0022-3514.63.1.17

Eveland Jr, William P., and Myiah H. Hively. 2009. "Political discussion frequency, network

size, and “heterogeneity” of discussion as predictors of political knowledge and

participation." Journal of Communication, 59 (2): 205–224.

Feldman, Lauren. 2011. "The Opinion Factor: The Effects of Opinionated News on Information

Processing and Attitude Change." Political Communication 28 (2): 163-181. doi: LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 41

10.1080/10584609.2011.565014

Garrett, R. Kelly. 2009. "Politically Motivated Reinforcement Seeking: Reframing the Selective

Exposure Debate." Journal of Communication 59: 676–699. doi:10.1111/j.1460-

2466.2009.01452.x

Gentzkow, Matthew, and Jesse M. Shapiro. 2006. "Media Bias and Reputation." Journal of

Political Economy 114 (2): 280-316.

Gentzkow, Matthew, and Jesse M. Shapiro. 2011. "Ideological Segregation Online and Offline."

Quarterly Journal of Economics, 126 (4): 1799–1839. doi:10.1093/qje/qjr044

Gerber, Alan S., Dean Karlan, and Daniel Bergan. 2009. "Does the Media Matter? A Field

Experiment Measuring the Effect of Newspapers on Voting Behavior and Political

Opinions." American Economic Journal: Applied Economics 1 (2): 35–52.

doi:10.1257/app.1.2.35

Groseclose, Tim, & Jeffrey Milyo. 2005. "A Measure of Media Bias." The Quarterly Journal of

Economics 120 (4): 1191–1237.

Gutmann, Amy, & Dennis F. Thompson. 1996. Democracy and disagreement. Harvard

University Press.

Hallin, Daniel. C., and Paolo Mancini. 2004. Comparing media systems: Three models of media

and politics. Cambridge University Press.

Holbert, R. Lance, R. Kelly Garrett, and Laurel S. Gleason. 2010. "A New Era of Minimal

Effects? A Response to Bennett and Iyengar." Journal of Communication 60 (1): 15–34.

Huckfeldt, Robert, Jeanette M. Mendez, and Tracy Osborn. 2004. "Disagreement, Ambivalence,

and Engagement: The Political Consequences of Heterogenous Networks." Political

Psychology 25: 65–95. doi:10.1111/j.1467-9221.2004.00357.x LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 42

Huckfeldt, Robert, & John Sprague. 1987. "Networks in Context: The Social Flow of Political

Information." American Political Science Review 81 (4): 1197–1216.

Iyengar, Shanto, and Kyu S. Hahn. 2009. "Red media, blue media: Evidence of ideological

selectivity in media use." Journal of Communication 59 (1): 19–39.

Iyengar, Shanto, Kyu S. Hahn, John A. Krosnick, and John Walker. 2008. "Selective exposure to

campaign communication: The role of anticipated agreement and issue public

membership." The Journal of Politics 70 (01): 186–200.

Jamieson, Kathleen Hall, & Cappella, Joseph N. 2008. Echo chamber: Rush Limbaugh and the

conservative media establishment. New York: Oxford University Press.

Johnson, Martin, & Kevin Arceneaux. 2011. "Does Media Fragmentation Produce Mass

Polarization? Selective Exposure and a New Era of Minimal Effects."

Kam, Cindy D., and Carl L. Palmer. 2008. “Reconsidering the Effects of Education on Political

Participation.” The Journal of Politics 70 (03): 612-631.

doi:10.1017/S0022381608080651.

Ladd, Jonathan McDonald, and Gabriel S. Lenz. 2009. “Exploiting a Rare Communication Shift

to Document the Persuasive Power of the News Media.” American Journal of Political

Science 53 (2) (January 1): 394–410.

Lazarsfeld, Paul F., Bernard R. Berelson, and Hazel Gaudet. 1948. The people’s choice: How the

voter makes up his mind in a presidential campaign. New York: Columbia University

Press.

Leighley, Jan E. 1990. "Social interaction and contextual influences on political participation."

American Politics Research 18 (4): 459–475.

Levendusky, Matthew S. 2011. "Rethinking the Role of Political Information." Public Opinion LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 43

Quarterly 75 (1): 42-64.

Levendusky, Matthew S. 2012. Partisan News That Matters. Book Manuscript.

McClurg, Scott D. 2006. "The electoral relevance of political talk: Examining disagreement and

expertise effects in social networks on political participation." American Journal of

Political Science 50 (3): 737–754.

Meffert, Michael F., Sungeun Chung, Amber L. Joiner, Leah Waks, L., and Jennifer Garst. 2006.

"The Effects of Negativity and Motivated Information Processing During a Political

Campaign." Journal of Communication 56 (1): 27–51. doi:10.1111/j.1460-

2466.2006.00003.x

Meredith, Marc, and Neil Malhotra. 2011. “Convenience Voting Can Affect Election

Outcomes.” Election Law Journal 10 (3): 227–253.

Mullainathan, Sendhil, & Andrei Shleifer. 2005. "The market for news." American Economic

Review 95 (4): 1031–1053.

Mutz, Diana C. 2002. "Cross-Cutting Social Networks: Testing Democratic Theory in Practice."

American Political Science Review 96 (01): 111–126. doi:10.1017/S0003055402004264

Mutz, Diana C. 2006. Hearing the other side: Deliberative versus participatory democracy.

Cambridge: Cambridge University Press.

Mutz, Diana C. 2008. "Is Deliberative Democracy a Falsifiable Theory?" Annual Review of

Political Science 11 (1): 521–538. doi:10.1146/annurev.polisci.11.081306.070308

Nir, Lilach. 2005. "Ambivalent social networks and their consequences for participation."

International Journal of Public Opinion Research 17 (4): 422–442.

Nir, Lilach, and James N. Druckman. 2008. "Campaign mixed-message flows and timing of vote

decision." International Journal of Public Opinion Research 20 (3): 326–346. LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 44

Prior, Markus. 2007. Post-broadcast democracy: How media choice increases inequality in

political involvement and polarizes elections. New York: Cambridge University Press.

Puglisi, Riccardo, and James M. Snyder. 2011. "Newspaper Coverage of Political Scandals."

Journal of Politics, 73 (3): 931–951. doi:10.1017/S0022381611000569

Scheufele, Dietram A., Bruce W. Hardy, Dominique Brossard, Israel S. Waismel-Manor, and

Erik Nisbet. 2006. "Democracy based on difference: Examining the links between

structural heterogeneity, heterogeneity of discussion networks, and democratic

citizenship." Journal of Communication 56 (4): 728–753.

Schudson, Michael. 1992. Was there ever a public sphere? If so, when?: reflections on the

American case.

Schudson, Michael. 1998. The good citizen: A history of American civic life. New York: Martin

Kessler Books.

Schudson, Michael. 2001. "The Objectivity Norm in American Journalism*." Journalism 2 (2):

149–170. doi:10.1177/146488490100200201

Stroud, Natalie J. 2006. "Selective exposure to partisan information." Ph.D. diss. University of

Pennsylvania.

Stroud, Natalie J. 2007. "Media effects, selective exposure, and Fahrenheit 9/11." Political

Communication 24 (4): 415–432.

Stroud, Natalie J. 2010. "Polarization and Partisan Selective Exposure." Journal of

Communication 60 (3): 556–576.

Stroud, Natalie J. 2011. Niche News: The Politics of News Choice. Oxford University Press.

Sunstein, Cass R. 2009. Going to extremes: how like minds unite and divide. Oxford University

Press. LOUD AND CLEAR: EFFECTS OF PARTISAN MEDIA 45

Taber, Charles S., and Milton Lodge. 2006. "Motivated skepticism in the evaluation of political

beliefs." American Journal of Political Science 50(3): 755–769.

Vallone, Robert, Lee Ross, and Mark Lepper. 1985. "The Hostile Media Phenomenon: Biased

Perception and Perceptions of Media Bias in Coverage of the Beirut Massacre." Journal

of Personality and Social Psychology 49 (3): 577-585.

Webster, James G. (2005). "Beneath the veneer of fragmentation: Television audience

polarization in a multichannel world." Journal of Communication 55 (2): 366–382.

Williams, Bruce A, and Michael X. Delli Carpini. 2002. "Heeeeeeeeeeeere’s Democracy!" The

Chronicle Review. April.

Williams, Bruce A., and Michael X. Delli Carpini. 2011. After Broadcast News: Media Regimes,

Democracy, and the New Information Environment. Cambridge: Cambridge University

Press.

Yanovitzky, Itzhak, and Joseph N. Cappella. 2001. "Effect of Call‐In Political Talk Radio Shows

on Their Audiences: Evidence from a Multi‐Wave Panel Analysis." International Journal

of Public Opinion Research 13 (4): 377–397. doi:10.1093/ijpor/13.4.377

LOUD AND CLEAR: SUPPORTING INFORMATION 46

SI.1 Programs used to estimate the partisan news score Cable news programs Anderson Cooper 360 (CNN) Hardball with Chris Matthews (MSNBC) The Beltway Boys (FNC) Heartland with John Kasich (FNC) The Big Story with John Gibson and Heather Larry King Live (CNN) Nauert (FNC) Late Edition with Wolf Blitzer (CNN) Countdown with Keith Olbermann (MSNBC) Lou Dobbs (CNN) CNN Headline News MSNBC Live Fox & Friends (FNC) Out in the Open (CNN) FOX News (FNC) Reliable Sources (CNN) The Fox Report with Shephard Smith (FNC) Situation Room (CNN) Geraldo at Large (FNC) Special Report with Brit Hume (FNC) Hannity & Colmes (FNC) Studio B with Shepard Smith (FNC) Hannity’s America (FNC) The O’Reilly Factor (FNC) Your World with Neil Cavuto (FNC) Network news programs 20-20 (ABC) Dateline NBC 60 Minutes (CBS) Face the Nation (CBS) Nightline (ABC) Frontline (PBS) ABC World News The McLaughlin Group America This Morning (ABC) Meet the Press (NBC) CBS Evening News NBC Nightly News CBS Morning News News Hour with Jim Lehrer (PBS) CBS Sunday Morning This Week with George Stephanopoulos (ABC) Non-news programs 24 Good Morning America (ABC) Big Love Law and Order Brothers and Sisters The Late Show with David Letterman The Colbert Report Oprah The Daily Show with Jon Stewart Scrubs CSI: Miami The Simpsons (FOX) The Early Show (CBS) The Today Show (NBC) Ellen DeGeneres Show The Tonight Show with Jay Leno Family Guy The View (ABC) Political talk radio Bill Bennett’s Morning in America The Mark Levin Show Bill O’Reilly Radio Factor Michael Savage, “The Savage Nation” The Clark Howard Show The Mike Gallagher Show The Glenn Beck Program The Neal Boortz Show The Jerry Doyle Show NPR’s All Things Considered The Jim Bohannon Show The Rush Limbaugh Show Dr. Laura Schlessinger The Sean Hannity Show Laura Ingraham

LOUD AND CLEAR: SUPPORTING INFORMATION 47

SI.2A Predicting Preference for McCain using Media Exposure Program b (SE) % Preferring McCain in Wave 2 NBC Nightly News .036 .288 (.082) The Simpsons .0010 .248 (.125) CBS Evening News -.0390 .270 (.087) CBS Morning News .0190 .265 (.126) CNN Headline News -.152 .260 (.090) The Tonight Show (Leno) .128 .332 (.090) America This Morning .1010 .262 (.249) The Daily Show with Jon Stewart -.639* .133 (.149) ABC News: Nightline -.289* .249 (.104) ABC World News .042 .289 (.084) Today Show -.002 .289 (.089) News Hour with Jim Lehrer -.339* .176 (.133) FOX News .514* .468 (.079) Good Morning America .070 .295 (.096) Ellen DeGeneres Show -.101 .214 (.106) 60 Minutes -.193* .271 (.084) Hannity & Colmes .360* .686 (.158) Hannity’s America .034 .830 (.256) Hardball with Chris Matthews -.205 .244 (.138) Heartland with John Kasich -.405 .769 (.385) 24 .004 .355 (.107) Late Edition with Wolf Blitzer -.246 .219 (.177) LOUD AND CLEAR: SUPPORTING INFORMATION 48

SI.2A, continued – Program b (SE) % Preferring McCain in Wave 2 The Late Show with David Letterman -.167 .272 (.092) The O’Reilly Factor .485* .618 (.137) The Early Show .056 .279 (.139) Face the Nation .130 .242 (.132) Fox & Friends .012 .606 (.127) Brothers and Sisters -.199 .227 (.112) Frontline -.207 .189 (.127) Oprah -.597* .236 (.091) Meet the Press -.425* .248 (.114) CBS Sunday Morning .128 .250 (.100) The Beltway Boys .283 .726 (.225) The Big Story with John Gibson & .214 .769 Heather Nauert (.283) 20-20 .225* .285 (.091) The Fox Report with Shepard Smith .250 .665 (.163) MSNBC Live .042 .234 (.111) CSI:Miami .131 .341 (.080) Out in the Open .080 .200 (.688) Law & Order -.194* .302 (.076) Situation Room -.193 .190 (.166) Special Report with Brit Hume .477 .659 (.185) Larry King Live -.118 .252 (.119) Lou Dobbs -.340* .250 (.160)

LOUD AND CLEAR: SUPPORTING INFORMATION 49

SI.2A, continued – Program b (SE) % Preferring McCain in Wave 2 Big Love -.037 .226 (.175) Dateline NBC .012 .288 (.087) Family Guy -.298* .249 (.118) Studio B with Shepard Smith -.030 .692 (.214) Reliable Sources .043 .225 (.397) Your World with Neil Cavuto .242 .687 (.185) McLaughlin Group .116 .259 (.134) The Week with George -.266* .211 Stephanopolous (.125) The View -.209 .216 (.118) Scrubs -.107 .269 (.105) The Colbert Report -.270 .167 (.153) Anderson Cooper -.127 .229 (.125) Geraldo at Large -.146 .534 (.190) Countdown with Keith Olbermann -.772* .131 (.171) Bill O’Reilly Radio Factor .311* .656 (.140) The Rush Limbaugh Show .896* .703 (.111) The Jim Bohannon Show -.458 .478 (.331) The Clark Howard Show .490* .583 (.202) The Mark Levin Show .102 .790 (.279) Bill Bennett’s “Morning in America” .278 .634 (.383) The Jerry Doyle Show .013 .622 (.511) NPR’s All Things Considered -.677* .197 .088

LOUD AND CLEAR: SUPPORTING INFORMATION 50

SI.2, continued – Program b (SE) % Preferring McCain in Wave 2 The Sean Hannity Show .444* .749 (.138) Michael Savage’s “Savage Nation” .377 .638 (.157) The Glenn Beck Program .622 .667 (.149) Dr. Laura Schlessigner .075 .545 (.157) Laura Ingraham .300 .759 (.216) The Neal Boortz Show .321 .693 (.227) The Mike Gallagher Show .023 .753 (.378) Constant -.084 (.068) Observations 5251 Pseudo R2 .255

LOUD AND CLEAR: SUPPORTING INFORMATION 51

SI.2B Comparing NAES and PEW 2008 Media Consumption Data

To assess the validity of our data, we break down the people who indicated that they watch each show by political ideology and compare to the proportions found for similar media questions in the 2008 Pew Biennial Media Consumption Survey. Because Pew 2008 often asked more general media consumption questions (usually identifying networks rather than specific shows), some comparisons required generation of comparable values from the show- specific answers provided in the NAES. Since the Media Partisanship Score method uses a binary (McCain-Obama) construction to generate MPS values, media variables are compared in terms of the proportion of self-identified (non-leaning) Democrats and Republicans. Pew responses are weighted as recommended in the codebook.

Notes: Asterisks denote shows for which there was not an exact match between the NAES and Pew Datasets. In these instances, NAES show-level responses were re-coded to match the Pew question as closely as possible. Shows / categories that showed significant differences in the Dem/Rep breakdown between the NAES and Pew respondents are in bold. Pearson’s X2 Program Survey % Dems % Reps (p, df=1) National Nightly NAES .590 .410 17.321 Network News* Pew .628 .372 (.000)

Cable News NAES .513 .487 35.323 Networks* Pew .556 .444 (.000)

CBS Evening NAES .611 .389 10.366 News Pew .705 .295 (.001)

ABC World News NAES .606 .394 16.945 Pew .710 .290 (.000)

NBC Nightly NAES .589 .411 8.266 News Pew .660 .340 (.004)

CNN* NAES .625 .375 1.923 Pew .654 .346 (.166)

Fox News Cable NAES .393 .607 47.086 Channel* Pew .541 .459 (.000)

MSNBC* NAES .674 .326 1.575 Pew .644 .356 (.209)

NPR NAES .713 .287 .775 Pew .691 .308 (.379) SI.2B, Continued: Comparing NAES and PEW 2008 Media Consumption Data Program Survey % Dems % Reps Pearson’s X2 LOUD AND CLEAR: SUPPORTING INFORMATION 52

News Magazine NAES .594 .406 8.323 Shows* (Dateline, 20/20) Pew .639 .361 (.004)

The NewsHour NAES .752 .248 2.013 with Jim Lehrer Pew .714 .286 (.156)

Late Night TV NAES .594 .406 1.704 Shows* (Letterman, Leno) Pew .621 .379 (.192)

Morning Talk NAES .585 .415 11.832 Shows* (Today Show, Pew .647 .353 (.001) GMA, The Early Show)

Sunday Morning NAES .628 .372 2.981 Talk Shows (Meet the Press, Pew .661 .334 (.084) This Week, Face the Nation)

Larry King Live NAES .659 .341 .317 (CNN) Pew .675 .325 (.573)

Countdown with NAES .826 .174 .312 Keith Olbermann (MSNBC) Pew .805 .195 (.576)

Hannity and NAES .167 .833 12.423 Colmes (FOX) Pew .265 .735 (.000)

The Daily Show NAES .810 .190 24.735 with Jon Stewart (Comedy Central) Pew .654 .346 (.000)

The Rush NAES .115 .885 22.131 Limbaugh Show (Talk Radio) Pew .243 .757 (.000)

SI.2B, Continued: Comparing NAES and PEW 2008 Media Consumption Data Program Survey % Dems % Reps Pearson’s X2 LOUD AND CLEAR: SUPPORTING INFORMATION 53

Lou Dobbs Tonight NAES .637 .363 .364 (CNN) Pew .662 .338 (.546)

Hardball with Chris NAES .687 .314 .205 Matthews (MSNBC) Pew .672 .328 (.651)

The O’Reilly NAES .218 .782 85.631 Factor (Fox) Pew .458 .542 (.000)

The Colbert Report NAES .753 .247 .418 (Comedy Central) Pew .774 .226 (.518)

LOUD AND CLEAR: SUPPORTING INFORMATION 54

SI.3 Lagged MPS Differences Predict Future MPS

To address the concern that within-subject differences in MPS represent noise rather than meaningful change, we also validated our measure using the approach employed by Levendusky (2011). Notably, there was significant variation in the MPS within individuals (p<.001 using a paired t-test), with the average person’s media diet becoming more homogenous and extreme during the general election (M=.488; SD=.308 during Wave 2; M=.526; SD=.304 during Wave 4; p<.001 using a paired t-test). To address the concern that these changes were due to measurement error, we estimated the following:

MPS i3 = β0 + β1(MPS i2 – MPS i1) + ui (3)

where the 1, 2, and 3 subscripts denote waves of the panel data1 and all other terms are defined as in Equations (1) and (2). As Levendusky (2011, p. 49) argues, finding a significant coefficient for the effect of change in MPS between Wave 2 and Wave 4 on the level of MPS in Wave 5 helps to rebut the argument that between-period differences are simply random noise stemming from measurement error. If this were true, change in MPS between Wave 2 and Wave 4 should have no predictive power to explain the level of MPS at Wave 5. The statistically significant β1 (p<.001) shown in Appendix 1 indicates that differences in the MPS reflect real change in media effects that persist over time. According to this coefficient estimate, people whose media diets become more (less) polarized and homogenous between the political primaries and general election receive higher (lower) MPS during the post-election period.

Lagged MPS Differences Predict Future MPS

Wave 5 MPS Wave 4 – Wave 2 MPS Differences .198* (.016) Constant .550* (.005) N 3487 R2 .031 * p < .05

______1 The careful reader will note that we are using three waves of panel data in order to validate our measure, whereas we are using two waves of data to test the relationship between media diet and polarization or participation. This is because data regarding both self-reported ideology and participation were only available in two waves subsequent to the media exposure variables. To reiterate, the participation variables were gathered during Wave 3 and Wave 5, and the ideology variables were gathered during Wave 3 and Wave 4. LOUD AND CLEAR: SUPPORTING INFORMATION 55

SI.4 Question Wording And Response Details For Polarization Measure

Iraq: “Which of the following plans for the United States Policy in Iraq comes closest to your own position?”

Response options: (1) The US should withdraw all troops from Iraq as soon as possible, regardless of conditions in Iraq. (2) The US should set a deadline for withdrawing its troops if the Iraqi government doesn't show definite progress in training Iraqi forces and controlling violence on its own. (3) The US should keep its troops in Iraq as long as is needed until a stable democratic government is established there.

In wave 2 (convention season), 52.7% of respondents indicated a polarized viewpoint (response 1 or 3). In wave 5 (the general election), 50.3% of respondents indicated a polarized viewpoint.

Path to Immigration: “Please indicate whether you favor or oppose each of the following proposals addressing immigration. Provide a path to citizenship for some illegal aliens who agree to return to their home country for a period of time and pay substantial fines.”

Response options: (1) Strongly favor. (2) Somewhat favor. (3) Somewhat oppose. (4) Strongly oppose.

In wave 2 (convention season), 33.2% of respondents indicated a polarized viewpoint (response 1 or 4). In wave 5 (the general election), 32.2% of respondents indicated a polarized viewpoint.

Border Security: “Please indicate whether you favor or oppose each of the following proposals addressing immigration. Increase border security by building a fence along part of the US border with Mexico.”

Response options: (1) Strongly favor. (2) Somewhat favor. (3) Somewhat oppose. (4) Strongly oppose.

In wave 2 (convention season), 57.0% of respondents indicated a polarized viewpoint (response 1 or 4). In wave 5 (the general election), 55.6% of respondents indicated a polarized viewpoint.

Free trade: “Do you favor or oppose the federal government in Washington negotiating more free trade agreements like NAFTA?”

Response options: (1) Strongly favor. (2) Somewhat favor. (3) Somewhat oppose. (4) Strongly oppose.

In wave 2 (convention season), 24.9% of respondents indicated a polarized viewpoint (response 1 or 4). In wave 5 (the general election), 24.7% of respondents indicated a polarized viewpoint.

LOUD AND CLEAR: SUPPORTING INFORMATION 56

SI.5 Variable Definitions and Coding Rules Used in Matching Analysis

This section details the coding rules used for the variables in the matching analysis. Variables like age, education, and income were recoded into groups (cohort, educational achievement, income tercile) to reduce the influence of small variations in these variables in the matching process.

Birth Cohort: based on age, respondents are coded into one of 6 cohorts: pre-New Deal (born 1914 or earlier), New Deal (born 1915-1930), post-New Deal (born 1931-1945), baby boomer (born 1946-1961), generation X (born 1962-1982), and generation Y (born 1983 or later).

Income Tercile: Income responses placed respondents into income ranges, which were recoded into relative income tercile for the matching analysis. The first tercile includes respondents indicating an income less than $50,000, the second includes respondents indicating an income greater than $49,999 and less than $85,000, and the third includes respondents indicating an income greater than or equal to $85,000.

Education: Respondents were placed into one of five groups based on their highest level of academic achievement: (1) less than a High School degree, (2) High School Degree (3) some College or Associate’s Degree, (4) Bachelor’s Degree, and (5) Graduate / Professional Degree.

Ideology: Respondents indicated their political ideology on a seven-point scale ranging from Extremely Liberal to Extremely Conservative.

Party Identification: Respondents indicated whether they were Democrats, Independents, or Republicans.

Bush Vote: 1 if the respondent voted for Bush in the 2004 election.

Political Interest: Respondents indicated their level of political interest on a four-point scale.

Media Partisanship Score (Wave 2): In order to examine the effect of changes in MPS over the course of the election, respondents were matched on their calculated MPS during the second wave of interviews.

Black: 1 if the respondent was Black, 0 otherwise.

Hispanic: 1 if the respondent was Hispanic, 0 otherwise.

Gender: 1 if the respondent was Male, 0 otherwise. LOUD AND CLEAR: SUPPORTING INFORMATION 57

TABLE SI.6 Difference in Means for Matched Variables Between Control and Treatment Groups, Treatment 1

Variable Unmatched Matched, No Caliper Matched, With Caliper Black .019 -.000 .000 (.010) (.009) (.015) Hispanic .015* .000 .000 (.007) (.007) (.005) Birth Cohort -.140* -.065* .000 (.028) (.023) (.061) Gender -.007 .004 .000 (.018) (.015) (.042) Income .061* .017 .000 (.030) (.025) (.072) Education .102* -.008 -.000 (.025) (.020) (.052) Ideology -.376* -.042 .000 (.058) (.051) (.121) Party ID -.187* -.014 .000 (.030) (.026) (.070) Bush Vote -.155* -.013 .000 (.013) (.015) (.039) MPS at Wave 2 .308* .110* .010 (.010) (.008) (.025) Political Interest .423* .048* .000 (.025) (.019) (.048)

N (Treatment 1 = 1) 2,167 2,167 261 N (Treatment 1 = 0) 1,030 1,030 1,030 Note: In column three, caliper = .25 Standard Deviations of the Mahalanobis Distance. Differences in bold are significant at the 5% level (two-sided), * p<.05. The table above shows the results for the “Political Participation” matching analysis, but results from the “Issue Polarization” matching analysis are functionally the same. LOUD AND CLEAR: SUPPORTING INFORMATION 58

TABLE SI.7 Difference in Means (Standard Deviation) for Matched Variables Between Control and Treatment Groups, Treatment 2

Variable Unmatched Matched, No Caliper Matched, With Caliper Black .017 -.000 .000 (.010) (.010) (.014) Hispanic .009 .000 -.000 (.007) (.007) (.005) Birth Cohort -.158* -.020 -.000 (.027) (.026) (.056) Gender -.001 -.005 .000 (.017) (.018) (.039) Income .058* .012 .000 (.028) (.029) (.070) Education .108* -.007 .000 (.023) (.022) (.046) Ideology -.472* -.012 .000 (.054) (.061) (.131) Party ID -.222* .012 -.000 (.028) (.031) (.071) Bush Vote -.183* -.004 -.000 (.017) (.017) (.039) MPS at Wave 2 .318* .078* .016 (.009) (.009) (.023) Political Interest .439* .012 -.000 (.023) (.020) (.040)

N (Treatment 1 = 1) 1,633 1,633 308 N (Treatment 1 = 0) 1,564 1,564 1,564 Note: In column three, caliper = .25 Standard Deviations of the Mahalanobis Distance. Differences in bold are significant at the 5% level (two-sided), * p<.05. The table above shows the results for the “Political Participation” matching analysis, but results from the “Issue Polarization” matching analysis are functionally the same. LOUD AND CLEAR: SUPPORTING INFORMATION 59

TABLE SI.8 Difference in Means (Standard Deviation) for Matched Variables Between Control and Treatment Groups, Treatment 3

Variable Unmatched Matched, No Caliper Matched, With Caliper Black .009 -.000 -.000 (.010) (.012) (.009) Hispanic .005 -.000 -.000 (.007) (.009) (.005) Birth Cohort -.157* -.018 -.000 (.028) (.032) (.059) Gender -.001 -.000 -.000 (.018) (.021) (.039) Income .056 .007 .000 (.030) (.035) (.067) Education .129* -.013 .000 (.024) (.027) (.048) Ideology -.482* -.001 -.000 (.057) (.075) (.145) Party ID -.248* .005 -.000 (.030) (.037) (.073) Bush Vote -.191* -.002 .000 (.018) (.021) (.039) MPS at Wave 2 .333* .071* .016 (.009) (.010) (.018) Political Interest .465* -.002 .000 (.025) (.022) (.032)

N (Treatment 1 = 1) 1,113 1,113 314 N (Treatment 1 = 0) 2,084 2,084 2,084

Note: In column three, caliper = .25 Standard Deviations of the Mahalanobis Distance. Differences in bold are significant at the 5% level (two-sided), * p<.05. The table above shows the results for the “Political Participation” matching analysis, but results from the “Issue Polarization” matching analysis are functionally the same.