Who can do my homework and accept the late payment? Simple vocabulary? read the description first

Midwest Political Science Association and Wiley are collaborating with JSTOR to digitize, preserve and extend access to American Journal of Political Science.

http://www.jstor.org

Midwest Political Science Association Wiley

Why Do Partisan Media Polarize Viewers? Author(s): Matthew S. Levendusky Source: American Journal of Political Science, Vol. 57, No. 3 (July 2013), pp. 611-623 Published by: Midwest Political Science Association Stable URL: http://www.jstor.org/stable/23496642 Accessed: 13-10-2015 05:37 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/ info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

This content downloaded from 140.211.95.10 on Tue, 13 Oct 2015 05:37:58 UTC All use subject to JSTOR Terms and Conditions

Why Do Partisan Media Polarize Viewers?

Matthew S. Levendusky University of Pennsylvania

The recent increase in partisan media has generated interest in whether such outlets polarize viewers. I draw on theories of

motivated reasoning to explain why partisan media polarize viewers, why these programs affect some viewers much more

strongly than others, and how long these effects endure. Using a series of original experiments, I find strong support for my

theoretical expectations, including the argument that these effects can still be detected several days postexposure. My results

demonstrate that partisan media polarize the electorate by taking relatively extreme citizens and making them even more

extreme. Though only a narrow segment of the public watches partisan media programs, partisan media's effects extend

much more broadly throughout the political arena.

America's

constitutional system, with its multi

ple veto points and separation of powers, re

quires compromise and consensus to function

effectively.1 Citizens can passionately advocate for their

beliefs, but they must also be willing to find a middle

ground if American government is to function effectively ( Gutmann and Thompson 2012). Many now claim, how

ever, that such compromise is increasingly out of reach in

American society, with deleterious consequences for our

politics (Gutmann and Thompson 2012). One potential

partial culprit for this lack of consensus is partisan media

outlets, such as Fox News. Such outlets provide view

ers with an "echo chamber" of their own beliefs, which

may in turn polarize them (Sunstein 2009). As citizens

move to the poles and harden their beliefs, it becomes

more challenging to find consensus solutions, and com

promise becomes more difficult and elusive. Do partisan media outlets bear part of the blame for the gridlock in

contemporary America? Do they make governance more

difficult? The article takes up these broad questions by asking

if partisan media polarize their audience, and if so, how

they do it. Drawing on theories of motivated reasoning, I offer a theoretical account of when and why partisan

media will polarize viewers. Previous work explains how

balanced sets of arguments can generate attitudinal po

larization (Taber, Cann, and Kucsova 2009; Taber and

Lodge 2006). I extend this logic to show how the unbal

anced presentation of the facts on partisan outlets will

generate even more marked levels of attitudinal polar

ization. I further demonstrate why these programs affect

some viewers much more strongly than others, and why

these polarizing effects endure, at least in the short term.

While I am not the first to suggest a link between

partisan media exposure and polarization (i.e., Stroud

2010; Sunstein 2009), previous work has relied on obser

vational data, rendering it unable to directly identify the

effects of media exposure. In contrast, I use a set of orig

inal experiments that allow me to isolate the impact of

partisan media on viewers' attitudes. Consistent with my

argument, I show that partisan media polarize viewers.

Further, these effects do not fade away immediately after

treatment, but can still be detected several days postexpo

sure. I also show that my experimental effects are largely

confined to a relatively small subset of the population:

those who actually watch partisan media programs in

the real world. My results show that these programs take

viewers who are already polarized and make them even

Matthew S. Levendusky is Assistant Professor of Political Science, University of Pennsylvania, 208 S. 37th St., Philadelphia, PA 19104

([email protected]).

The author thanks Daniella Lejitneker and the staff of the Wharton Behavioral Lab for helping to implement experiment 1 in the article.

I would like to thank Kevin Arceneaux, Paul Beck, Adam Berinsky, Robin Blom, John Bullock, Jamie Druckman, Bob Erikson, Stanley Feldman, John Gasper, Don Green, Greg Huber, Luke Keele, Jonathan Ladd, Gabriel Lenz, Neil Malhotra, Marc Meredith, Diana Mutz,

Jeremy Pope, Rogers Smith, Laura Stoker, Josh Tucker, the editor, anonymous referees, and many others for helpful comments. Seminar

participants at APSA 2011, Berkeley, MIT, Northwestern, NYU-CEES, and Stanford also gave very useful feedback. Any remaining errors

are my own.

'Data to replicate this study are available from the author's dataverse (http://dvn.iq.harvard.edu/dvn/dv/mleven). The School of Arts and

Sciences and the Vice Provost for Research at the University of Pennsylvania generously provided funding for this research.

American Journal of Political Science, Vol. 57, No. 3, July 2013, Pp. 611-623

©2013, Midwest Political Science Association DOI: 10.1111/ajps. 12008

6u

This content downloaded from 140.211.95.10 on Tue, 13 Oct 2015 05:37:58 UTC All use subject to JSTOR Terms and Conditions

612 MATTHEW S. LEVENDUSKY

more extreme. Partisan media therefore heighten mass

polarization not by turning moderates into extremists,

but rather by further polarizing those who are already

away from the political center.

This, in turn, illustrates that partisan media do con

tribute to the difficulty of consensus in American pol itics. While partisan media only reach a small segment of the public, because that segment is more extreme and

engaged, they have an outsized political role (Bai 2009;

Jamieson and Cappella 2008). By affecting a more engaged and influential segment of the mass public, partisan me

dia impact American politics quite broadly. Even though the audience for partisan media is quite small, its effects

on American politics are not.

Studying the Effects of Partisan Media

Scholars differentiate mainstream or detached media out

lets, which prize balance, fairness, and objectivity, from

partisan outlets, where stories are "framed, spun, and

slanted so that certain political agendas are advanced"

(Jamieson, Hardy, and Romer 2007, 26). Partisan media

are opinionated media: media that not only report the

news but offer a distinct point of view on it as well. One

could certainly characterize these shows as biased in favor

of one party and political viewpoint—they create a coher

ent liberal or conservative vision of the news. The hosts

package the news in a way to help people make sense of

the world, creating a "self-protective enclave" of consis

tent messages and a framework to understand the day's

events (Jamieson and Cappella 2008, χ). These shows also

engage in a biased story selection, reporting more heavily on topics that favor their sides and downplaying stories

that harm their points of view (Baum and Groeling 2008,

2010). To the extent they discuss the "other side," they do so in a straw-man fashion, one better suited for easy

dismissal than serious debate. Partisan news programs are not primarily about conveying facts; they are about

helping people make sense of the world given particular

predispositions (Rosensteil 2006, 253). These one-sided

messages give viewers an easily digestible version of an

otherwise confusing political world.

These partisan media programs are typically not,

however, the "shout shows" lamented for their incivil

ity (Mutz 2007). These shows are not about argument: the arguing is done, and one side has clearly won—at

least in the mind of the host (Rosensteil 2006). While the

clips of Bill O'Reilly shouting at Democratic Congress man Barney Frank are legendary, they are also relatively

rare. Agreement with the host's viewpoint is the norm

here.

One might be tempted to dismiss these programs as insignificant because their audience is relatively lim

ited and partisan, but this claim is shortsighted on sev

eral levels. First, at the most basic level, even if these

shows only reach a small segment of the market, this is

a deeply politically engaged audience, with many influ

ential citizens who will make their voices heard in the

halls of power (Bai 2009; Jamieson and Cappella 2008). In this instance, limited numbers do not mean limited

influence.

Further, the over-time data suggest that the audience

for partisan media programs has grown dramatically in

recent years. While it is true that the audience for parti san media shows is smaller than the audience for nightly network news, the audience for partisan shows is growing while the broadcast audience is shrinking (Project for Ex

cellence in Journalism 2009). For example, Fox News did

not exist until 1996, and only 20% of American munici

palities had Fox by the year 2000, but today nearly all mar

kets have access to the station—quite remarkable growth over a 15-year period (DellaVigna and Kaplan 2007). The over-time data on audience size suggest that these

kinds of shows and networks are likely to continue to

grow apace (Pew Center for the People and the Press

2010; Project for Excellence in Journalism 2009). These

over-time audience trends need to be understood in the

broader context of a postbroadcast media environment.

Gone are the days of one "mainstream" message com

ing from the major broadcast networks; today's media

environment is characterized by a proliferation of news

sources, many matched to viewers' partisan predisposi

tions (Prior 2007). Indeed, partisan media sources are

simply one example of this broader trend of seeking out

like-minded information, albeit a relatively extreme one

(Bennett and Iyengar 2008; Iyengar and Hahn 2009). Given this new reality, scholars need to understand its

consequences.

Many older media studies concluded that media ex

posure had a muted effect on attitudes (the "minimal

effects" hypothesis), but if anything, media exposure rein

forced attitudes (Berelson, Lazarsfeld, and McPhee 1954;

Klapper 1960). More recent scholarship, however, sug

gests that the media may have a stronger influence than

scholars initially suspected. In particular, media slant

influences attitudes and vote choice (i.e., Barker 1999;

Dalton, Beck, and Huckfeldt 1998; DellaVigna and Kap lan 2007). One might therefore expect partisan media

to polarize viewers, at least when they expose viewers

to like-minded, proattitudinal information (though see

Mutz 2006).

This content downloaded from 140.211.95.10 on Tue, 13 Oct 2015 05:37:58 UTC All use subject to JSTOR Terms and Conditions

WHY DO PARTISAN MEDIA POLARIZE VIEWERS? 613

But there are two important limitations to previous

research. First, drawing any sort of firm causal inference

from the existing literature is extremely difficult. View ers typically self-select into media sources, so it is very difficult to determine if media slant impacts voters or if viewers choose sources that reflect their preferences

(Mutz 2006). While the slant of an outlet impacts its au

dience, the audience's preferences also drive the outlet's

slant (Gentzkow and Shapiro 2010; Mullainathan and

Shleifer 2005). Previous studies relying on observational

data cannot differentiate selection from treatment effects.

Clearly determining the effect of exposure to a particular medium independent of the selection effect is extremely difficult, a point that becomes even more pronounced in

the contemporary high-choice media environment with

its larger range of choices.2

Further, and more importantly, scholars know little

about the mechanisms through which these shifts occur or

which viewers will be affected by the treatments. Are cer

tain types of partisan media more effective at polarizing viewers than others? Are certain viewers more respon

sive? For example, not all subjects will willingly watch

news programs, particularly partisan news programs, so

the effects of these shows will be very unevenly distributed

throughout the population. It is not simply enough to say that media can reinforce attitudes; we need to know un

der what circumstances, how, and for whom these effects

occur. Understanding these sorts of moderators is crucial

to advancing our understanding of the effects of partisan media.

How Partisan Media Polarize Viewers

I start from the premise that humans are motivated rea soners (Kunda 1990). Humans have two broad classes

of goals: accuracy goals (the desire to reach the correct

conclusion) and directional goals (the desire to reach the

preferred conclusion, i.e., the conclusion that supports

our existing beliefs). Human reasoning relies on both, but

directional goals have an especially strong effect: we pro

cess information so that it fits with our existing beliefs.

When citizens hear a news story about (say) President

Obama, simply upon hearing his name, their attitudes

and feelings toward him come to the fore, even without

any conscious thought (a process known as "hot cogni tion"; see Lodge and Taber 2005; Morris et al. 2003). These

2There are important rare exceptions, however (Ladd and Lenz

2009). There are also experimental designs that address self

selection (Arceneaux and Johnson 2010; Gaines and Kuklinski

2011), but that is a separate question from my focus here.

thoughts and feelings about the president then shape how citizens interpret the evidence provided in the story. If the information suggests President Obama is ably handling his job, supporters of the president will uncritically accept the news, while his critics will counterargue and challenge it (a process known as disconfirmation bias; see Ditto and

Lopez 1992). As a result, even balanced sets of arguments

can generate attitudinal polarization (Taber, Cann, and

Kucsova 2009; Taber and Lodge 2006). Partisan media programs intensify this motivated

reasoning because of their slanted presentation of the

news. Consider first how this process occurs when sub

jects watch like-minded media—that is, proattitudinal media that reinforce their existing beliefs, such as when a

conservative Republican watches Fox News. Such pro

grams heighten motivated reasoning for two reasons.

First, these programs broadcast one-sided, proattitudi nal messages to viewers, which they will uncritically ac

cept (Taber, Cann, and Kucsova 2009; Taber and Lodge 2006). But absent any competing message, this general

tendency to accept proattitudinal information becomes

even stronger: because it lacks a counterargument of any

sort, this information (implicitly) seems stronger and

even more persuasive (Klayman and Ha 1987; Lodge and

Taber 2001; Zaller 1996). Subjects will therefore move in

the direction of the evidence and become more extreme

(Moscovici and Zavalloni 1969). The type of "echo cham

ber" environment found in like-minded media will push viewers toward the ideological extremes, thereby polariz

ing their attitudes.

But there is another reason why like-minded content

will magnify the tendency toward attitudinal polarization. These shows present the day's news as a partisan struggle,

with clear references to the political parties and their po sitions. This primes citizens' partisanship, strengthening the degree to which they see the world through partisan

colored glasses (Campbell et al. [1960] 1980; Goren,

Federico, and Kittleson 2009; Price 1989). Priming this

sort of salient identity increases viewers' directional

goals—it heightens their desire to reach a conclusion in

line with their partisanship, thereby strengthening their

biases toward attitudinally congenial information. Cueing

partisanship increases its ability to slant how subjects see

the world by strengthening the desire to engage in the mo

tivated reasoning described above. This is consistent with

work in social psychology demonstrating that priming relevant group identities, such as partisanship, increases

attitudinal polarization (Abrams et al. 1990; Lee 2007). So while there is a general tendency toward attitudinal

polarization in political settings simply because humans

are motivated reasoners, the unique environment of like

minded partisan media—with its one-sided content and

This content downloaded from 140.211.95.10 on Tue, 13 Oct 2015 05:37:58 UTC All use subject to JSTOR Terms and Conditions

614 MATTHEW S. LEVEND USKY

partisan primes—should be especially likely to generate attitudinal polarization. This leads me to state my first

hypothesis:

HI: Exposure to like-minded partisan media will polarize attitudes (i.e., increase attitudinal extremity).

But what happens when subjects watch cross-cutting

media that present them with counterattitudinal mes

sages (i.e., when a Democrat watches Fox News)? Here,

subjects' preferences are challenged rather than re

inforced. When presented with information running counter to their prior beliefs, subjects will attempt to dis

credit and counterargue it (Ditto and Lopez 1992; Taber,

Cann, and Kucsova 2009; Taber and Lodge 2006). Sub

jects will therefore believe they have refuted this informa

tion, which makes the evidence supporting their own side

even more persuasive, pushing them toward greater atti

tude extremity (Redlawsk 2002; Taber and Lodge 2006).

Cross-cutting media therefore increase attitude extremity

and polarization, a "boomerang" effect.

H2: On average, cross-cutting media will polarize atti

tudes.3

While on average there should be a polarizing effect of

cross-cutting media, that effect should be especially pro nounced for a particular subset of the audience. Polariza

tion here stems from counterarguing these cross-cutting

messages, but not all viewers are equally equipped to do

this—subjects need to be informed about the issue, and

to care deeply about it, to effectively generate these coun

terarguments (Taber and Lodge 2006). The ability to do this comes from holding strong attitudes. Strong attitudes

are ones where the subjects know more about the issue,

and hence can more easily generate counterarguments,

and the issue is more central and important to them,

so they will be more motivated to generate counterargu

ments (Krosnick and Petty 1995). Strong attitudes there

fore help subjects polarize in response to cross-cutting media (Pomerantz, Chaiken, and Tordesillas 1995; Taber

and Lodge 2006; Zaller 1992).

H3: Exposure to cross-cutting media will especially in

crease attitude extremity for viewers with strong prior attitudes.

3I draw on theories of motivated reasoning here because they offer

the most complete and compelling account of this process. One

could also derive similar predictions from a variety of other theo

retical models, such as group polarization theories (Isenberg 1986) or theories of source credibility (Lupia and McCubbins 1998).

How Long Do These Effects Last?

One enduring concern with media persuasion studies is

that such effects rapidly fade away (Druckman and Nelson

2003). But given the theoretical mechanism driving these

partisan media effects, there is good reason to suspect that these effects endure, at least in the short term. While

citizens use both memory-based and online processing to

update their beliefs (Redlawsk 2001), online processing is more central when subjects are utilizing affect-laden

motivated reasoning processes (Taber and Lodge 2001).

So, for example, when citizens watch a partisan media

program about gun control, they bring to mind their

attitude toward gun control, update it on the spot using

the new information in the segment, and then store this

updated attitude in long-term memory (Lodge, McGraw, and Stroh 1991). Later on, viewers may not remember

the specific arguments from the partisan media host that

caused them to update their opinions, but they should

remember their overall attitude—the details fade away, but the summary attitude remains (Lodge, Steenbergen, and Brau 1995). Attitudes updated in response to partisan media, then, should not immediately fade away.

H4: The polarizing power of partisan media will endure

for several days postexposure.

Research Design

To test these hypotheses, I conducted a series of original experiments. All of the experiments follow one of two

basic protocols. In study 1, subjects were recruited to take

part in a study about how individuals learn about poli tics from the news; the study took place in the decision

making laboratory of a private urban university. After

reading some basic instructions, subjects completed a

pretest questionnaire to measure their baseline attitudes

and opinions. They then completed a brief distracter task

(a series of items from an IQ test), watched the stimulus

(see below), and then completed the posttest battery of

attitudinal items. Study 1 therefore allows me to measure

pretest attitudes and attitudinal strength to test Hypoth esis 3, as well as directly measure attitudinal change in

response to the stimuli.

In the remaining studies, subjects were recruited on

line to take part in a study of how individuals learn about

the news. In these experiments, subjects began by an

swering a few background items, then were exposed to the

stimulus, and finally completed the posttest attitudinal as

sessment (so these experiments are posttest-only designs,

This content downloaded from 140.211.95.10 on Tue, 13 Oct 2015 05:37:58 UTC All use subject to JSTOR Terms and Conditions

WHY DO PARTISAN MEDIA POLARIZE VIEWERS? 615

whereas experiment 1 is a pre-post design). While the pre

post design has the advantage that one can measure pretest

variables and directly measure within-person change, it

has the disadvantage that subjects might remember their

pretest answers when completing the posttest, even with

a distracter task in place. Using both the pre-post and

posttest-only designs ensures that my design does not

overly influence my results. Additional details on the ex

periments are given in the appendix. In all experiments, subjects were exposed to the same

type of stimulus: a series of video clips from like-minded,

cross-cutting, or neutral media programs (treatment as

signment is held constant for each subject in each experi ment) .These video clips came from actual news programs

and cover issues discussed during the first 18 months

of the Obama presidency.4 I selected like-minded and

cross-cutting shows by using external judgments about

the slant of their news coverage. For all subjects, neutral

clips came from the PBS News Hour, considered to be

one of the most balanced shows on TV (Groseclose and

Milyo 2005). For Republicans (Democrats), like-minded

(cross-cutting) programs came from Fox News (Han

nity and The O'Reilly Factor)·, cross-cutting (like-minded)

programs came from MSNBC ( Countdown with Keith Ol

bermann).5 I selected Fox News as the right-wing source

as it favors Republicans/conservatives (Jamieson and

Cappella 2008); likewise, I selected MSNBC as the left

wing source for parallel reasons (Steinberg 2007). The

particular programs selected are among the most popu

lar shows on each network, making them suitable choices

that subjects should perceive as highly realistic.

In each experiment, subjects across all treatment con

ditions hear information about the same issue(s); the

difference among conditions is the partisan bias of the

source. My manipulation, then, estimates the effects of

exposing subjects to partisan media (with a like-minded

or cross-cutting message), relative to a control baseline of

neutral media (which provide information without any

particular partisan bias).6 To ensure that subjects could detect the differences

in partisan bias across sources, each experiment included

4To ensure that the results do not hinge on a particular issue, each

experiment uses a slightly different mix of issues; see the appendix for the issues used in each experiment as well as a transcript of the

segments used.

5Partisan leaners are treated as partisans (Keith et al. 1992); pure

Independents (those who do not lean toward either party) are

dropped from the analysis. Including them does not change the

substantive results (see the appendix).

6An alternative specification would be to estimate the effects of

partisan media relative to an apolitical control condition; doing so would not change the substantive results reported here (see the

appendix for more discussion).

a manipulation check item asking subjects to assess the

partisan tilt of the segments they watched. In every ex

periment, subjects had no difficulty discerning the slant

of the sources (see the appendix), supporting my claim

that it is in fact the partisan biases that drive the results

below.

Given that these are real-world clips, and not man

ufactured stories, however, there may be some unmea

sured difference between the sources that could affect the

results in some manner. I use real-world clips to gain

greater verisimilitude, even at the expense of having less

control over the sources. But to ensure that this decision

did not bias my results, I conducted a follow-up study where I used manufactured newspaper editorials as the

treatment stimuli. Here, I gain greater control over con

tent (and more internal validity) at the expense of realism

(and hence lower external validity). The results from this

follow-up experiment replicate the main findings dis

cussed here, which should bolster the reader's confidence

in the results I report below (see the appendix for details). Note that this design also allows me to synergize pre

vious related work. Taber and Lodge (2006) study a similar

theoretical process, but in a more stripped-down, abstract

experimental context. On the other hand, Stroud (2010) studies the potential polarizing power of partisan me

dia, but using real-world observational data. By using an

experiment featuring real-world partisan media clips, I

combine the strengths of both approaches. My findings are valuable in their own right but also offer a unique

perspective on these earlier results.

Data and Analyses

To examine whether partisan media polarize viewers, each posttest includes a series of questions designed to

elicit respondents' attitudes. To measure polarization, I

look specifically at attitude magnitude, with larger mag nitudes indicating greater polarization: respondents are

more polarized when they "strongly agree" with a policy versus when they "agree" with it. The items asked in each

posttest focus on the issues discussed in the stimuli: so,

for example, if the stimuli focused on the Bush tax cuts, then subjects answered items about the Bush tax cuts.

This means that my results here can only speak to issue

polarization (i.e., changes in polarization related to the

specific issues discussed on the segments), rather than

more general ideological polarization, which I leave for

future work.

Each experiment includes a battery of items which are

combined to form an index of opinion on the issue that

serves as the dependent variable in the analyses below;

This content downloaded from 140.211.95.10 on Tue, 13 Oct 2015 05:37:58 UTC All use subject to JSTOR Terms and Conditions

6l6 MATTHEW S. LEVENDUSKY

see the appendix for the specific items used and the scale

construction. To simplify the interpretation of the results

presented below, I rescale the dependent variable in each

model to lie in the [ — 1,1] interval. This makes it easier

to intuitively understand the treatment effects in each re

gression, since they are reported on a common scale.7 I

analyze folded versions of the dependent variable, so that

higher values indicate stronger agreement with a sub

ject's party's position (i.e., Democrats with higher scores

take more liberal positions, and Republicans with higher scores take more conservative positions). This folding as

sumes that the process is equivalent across parties (e.g., that the mechanisms work the same way for Democrats

and Republicans). I enforce this constraint here because

a more flexible model (where the effect differed by party) found no significant differences by party (see the ap

pendix for more details and results). I begin by testing the effects of like-minded and cross

cutting media on attitudinal polarization using the data

from experiment 1.1 estimate:

(yfT -

rty) = βο + βι LMi + fccc,

+ β}Ζί + Φ; + φj + Ujj,

where yf" is respondent is pretest attitude on issue j,

yf°st is the equivalent posttest attitude, LM,· and CC, are

indicators for whether respondent i was assigned to the

like-minded or cross-cutting conditions (respectively), Z; is a vector of demographic control variables, φ are

the respondent-specific random effects, the φ are issue

specific fixed effects, and u is a stochastic disturbance

term. I include the respondent random effects and issue

fixed effects because experiment 1 exposes subjects to in formation about four different issues, so the data from

experiment 1 are in a panel format.8 As an additional

guard against heterogeneity, I also cluster the standard

errors by respondent in the analysis reported below.9

If Hypotheses 1 and 2 are correct, I should find that

like-minded media exposure and cross-cutting media ex

posure both have a positive and significant coefficient,

indicating that, on average, those assigned to watch like

minded and cross-cutting media become more extreme.

7 For descriptive statistics on the dependent variable from each

experiment, see the appendix.

8To be clear, in experiment 1, the stimulus discusses four different

issues, so there are four issue-specific scales (each yis a scale

specific to a particular issue). I analyze the data in a panel format

for simplicity, since my focus is on the average effect across issue

rather than the results for any particular issue. I note, however, that

I would reach the same conclusion if I analyzed the items separately

by issue.

'Omitting these subject/item effects (or clustering) would not

change my substantive conclusions reported here.

Table 1 Effects of Partisan Media on Attitude

Extremity, Experiment 1

Variable (1) (2) (3) (4)

Like-Minded 0.14 0.13 0.14 0.13

Treatment (0.04) (0.04) (0.04) (0.04)

Cross-Cutting -0.03 -0.03 -0.05 -0.05

Treatment (0.04) (0.04) (0.04) (0.04)

Strong Attitude -0.04 -0.04

(0.06) (0.06) Like-Minded 0.03 0.03

Treatment* Strong (0.08) (0.09) Attitude

Cross-Cutting 0.22 0.23

Treatment* Strong (0.08) (0.09) Attitude

Partisanship 0.01 0.01

(0.01) (0.01) Male 0.00 0.01

(0.03) (0.03) Student 0.08 0.07

(0.05) (0.05) White -0.04 -0.03

(0.03) (0.03)

Age 0.01 0.01

(0.00) (0.00)

Intercept -0.06 -0.32 -0.06 -0.31

(0.04) (0.11) (0.04) (0.12) N 720 720 720 720

Number of Subjects 164 164 164 164

R-Squared 0.05 0.06 0.05 0.06

Note: Cell entries are OLS regression coefficients based on the

model above with robust standard errors in parentheses (fixed ef

fects for issue and random effects for respondents are included in

all models but are not reported). Coefficients that can be distin

guished from 0 (a < 0.10, one-tailed) are given in bold. Note that

the excluded baseline group in the analysis is subjects assigned to

the neutral (PBS) condition. For details on variable construction, see the appendix.

Table 1 presents the results. The results strongly support

Hypothesis 1: like-minded media make subjects more ex

treme (relative to subjects assigned to the neutral media

condition). Following like-minded media exposure, sub

jects become approximately 8% more extreme. Exposure to like-minded media increases polarization.

Note, however, that in columns 1 and 2 in Table 1

there are no effects for cross-cutting media—the effect is

statistically insignificant and close to 0, indicating that

voters simply seem to "tune out" cross-cutting mes

sages. But perhaps the effect of cross-cutting media is

heterogeneous, with viewers with strong attitudes more

This content downloaded from 140.211.95.10 on Tue, 13 Oct 2015 05:37:58 UTC All use subject to JSTOR Terms and Conditions

WHY DO PARTISAN MEDIA POLARIZE VIEWERS? 617

likely to polarize in response to this content, as I ar

gued in Hypothesis 3. Column 3 of Table 1 tests Hy

pothesis 3, interacting pretest attitude strength with treat

ment assignment. The results here support this hypoth

esis. While the overall effect of cross-cutting media is

not statistically significant, the interactive effect of cross

cutting media and strong pretest attitudes is significant.10

Cross-cutting media have the potential to polarize view

ers, but not all seize that opportunity: only those with

strong attitudes actually polarize in response to these

messages.11

In contrast, there is no interactive effect of attitude

strength for like-minded media. This fits with the theo

retical mechanism offered above: cross-cutting media in

crease extremity only among those with strong attitudes

because those are the respondents with the cognitive re

sources to resist persuasion. In contrast, for like-minded

media, attitude strength is irrelevant, because there is no

need to counterargue the message. Given this, the lack of

a moderating effect of attitude strength on like-minded

media should not be surprising.

Are These Results Externally Valid?

The results from experiment 1 demonstrate that partisan

media can polarize viewers. But because humans are mo

tivated reasoners, one might be skeptical of these results.

Motivated reasoning not only predicts how subjects will

respond to new information (as I argued above), but it

also predicts the type of information subjects will prefer. Motivated reasoning leads subjects to seek out more like

minded content both generally (Druckman, Fein, and

Leeper 2012; Taber and Lodge 2006) and specifically in the

10Here, attitude strength is measured as an additive index of atti

tude certainty, importance, and (self-perceived) knowledge about

the issue; see the appendix for more details. To simplify the in

terpretation of the results, I dichotomized strength in the analysis in Table 1, counting those in the top decile as having "strong" at

titudes. One can obtain similar, albeit weaker, results using those

in the top half of strength, suggesting that much of the effect is

concentrated among those with the "strongest" attitudes.

"One concern with these sorts of moderator effects, however, is

that they might be picking up other differences between subjects: for example, maybe those with stronger attitudes are simply more

interested in politics and hence pay closer attention to the stimuli

and the resulting questions, generating the effects seen in the text.

I can never completely rule out this possibility, but I can offer a

test of the underlying idea. I can interact the treatment with both

attitude strength and a measure of political interest/attentiveness.

If attitude strength is just a proxy for general interest/engagement, then the attitude strength results should go away once I include

the interaction with the interest/engagement measure. When I do

this, my results survive, suggesting that strength (and not just in

terest/engagement or some other factor) drives these results.

case of partisan media (Iyengar and Hahn 2009; Stroud

2011). Therefore, the sort of forced assignment design used here might potentially give misleading results: be

cause of motivated reasoning, subjects prefer like-minded

content, so it is unclear how experiment 1 maps into the

real world.

This is potentially a serious challenge to my results:

as experiments, they have high internal validity, but they

might have limited external validity precisely because of

my proposed theoretical mechanism. To help address this

point, I conducted what is known as a patient preference

trial (Torgerson and Sibbald 1998). This is a standard

random-assignment experiment with one small twist. In

the pretest, I solicit subjects' preferred treatment: if they had to choose, would they prefer a like-minded, cross

cutting, or neutral news source.12 I can then condition

on this preference to examine if those who would ac

tually want to watch these shows respond differently to

them than other respondents. This allows me to deter

mine whether my results have at least some external va

lidity: do those who actually want to watch these shows

respond to them as my theory predicts? Or does showing

respondents a segment they would never actually watch in

the real world drive the earlier results? While this analysis has slightly lower internal validity (given that it involves

conditioning on an observed variable), its external valid

ity nicely complements experiment 1.

But for this approach to succeed, I need to know that

my preference measure is actually a valid way to estimate

real-world viewing habits. To do so, I conducted a follow

up study where I asked subjects about their preferences for

different types of media, as well as their actual media con

sumption. I find that viewer preferences are very strongly related to actual media consumption. For example, 48%

of those who prefer like-minded media watch it daily ver

sus 5% for those who prefer another type of media. On

average, those who prefer like-minded media watch it at

least weekly versus only a few times a year for those who

prefer other types of media (see the appendix for more

details). Preferences actually measure viewing habits, and

my measure therefore allows me to examine how viewing

habits condition the effect of the treatment.13

l2To make the item as easy to understand as possible, subjects are

asked if they would prefer a show from Fox News, MSNBC, or

PBS, with appropriate examples from each network (i.e., the labels

"like-minded," etc., are not used); see the appendix for the specific

question wording.

13My setup only solicits preferences over political media, not apolit ical media. My work establishes what happens in a political media

context, but an important next step will be to consider what hap

pens when apolitical media are added to the mix (see Arceneaux

and Johnson 2010).

This content downloaded from 140.211.95.10 on Tue, 13 Oct 2015 05:37:58 UTC All use subject to JSTOR Terms and Conditions

6l8 MATTHEW S. LEVENDUSKY

These preferences should strongly shape how viewers

respond to the treatment. I expect that much of the po

larization in response to like-minded media will be con

centrated among those who prefer to watch like-minded

programs. Citizens with stronger directional goals and

more cognitive resources are more likely to demonstrate

the biased processing characteristic of motivated reason

ers (Taber and Lodge 2006). Both of these descriptions—

strong directional goals and highly cognitively skilled—

apply to those who prefer like-minded content.

Like-minded media viewers are more strongly partisan and ideological (Baum and Groeling 2010; Jamieson and

Cappella 2008; Stroud 2011) and hence have greater incentives to react strongly to the partisan framing of

these programs, resulting in stronger directional goals. Further, these subjects also possess more of the cognitive

skills needed to process these sorts of messages: they

are more politically engaged, better informed, more

active in politics, and so forth (Baum and Groeling 2010;

Jamieson and Cappella 2008; Stroud 2011; see also the

appendix). Citizens who choose to watch like-minded

programs will therefore be more likely to engage in

motivated reasoning and hence will be more likely to

polarize in response to like-minded content. Put slightly

differently, like-minded media should primarily polarize those who want to watch it.

In contrast, I expect those who prefer cross-cutting

media to become less polarized in response to cross

cutting messages. These individuals are quite out of step with their parties, more likely to defect from them, and

feel psychologically closer to the opposing party—the de

sire to consume cross-cutting media signals that they are

"odd" partisans (Holbert, Garrett, and Gleason 2010).

They therefore have more motivation to respond to the

cross-cutting cues. As a result, I expect them to move to

ward the cross-cutting source's position and thereby hold

more moderate (less polarized) attitudes.14 Subjects' re

action to other types of programs (i.e., how those who

prefer like-minded media will respond to cross-cutting media) is less clear, and I correspondingly have no ex ante

directional expectations in these cases.

Table 2 estimates how these preferences impact the

treatment effects.15 The analysis here parallels the analysis of experiment 1, but I interact treatment assignment and

these preferences:

14An important extension for future work, however, will be to more

carefully consider which subjects opt into consuming cross-cutting media and why they choose to do so. Doing so will require a more

general theory of cross-cutting media.

15In the data from experiment 2 used here, 32% of subjects prefer to watch like-minded media (versus 18% for cross-cutting media

and 50% for neutral media).

Table 2 The Role of Viewer Preferences,

Experiment 2

Variable Estimate

Like-Minded Treatment -0.05

(0.12)

Cross-Cutting Treatment 0.08

(0.12) Prefer Like-Minded Media -0.07

(0.14) Prefer Cross-Cutting Media -0.22

(0.18) Like-Minded Treatment* Prefer 0.28

Like-Minded Media (0.19) Like-Minded Treatment* Prefer -0.17

Cross-Cutting Media (0.26)

Cross-Cutting Treatment*Prefer -0.08

Like-Minded Media (0.20)

Cross-Cutting Treatment* Prefer -0.31

Cross-Cutting Media (0.23) Constant 0.21

(0.08) N 163

R-Squared 0.14

Note: Cell entries are OLS coefficients with associated standard

errors; see Table 1 for additional details.

Yi — βο + βι LMj + frCQ + β; Ρ L t

+ β4 PC; + βδΙΜ, * Ρ Lj + ftfrCCi * Ρ Li

-f- β7 L Mi * Ρ Ci fyC Ci * Ρ Ci -f~ //;,

where y is respondent ΐs posttest attitude extremity, PL

and PC are indictors if respondent i prefers like-minded

or cross-cutting media (respectively), and all other terms

are as defined above. Note that in experiment 2 (and

experiment 3 below), subjects are only asked about one

issue, so there is no need for a panel analysis as in ex

periment 1. In the analysis of model (2), interest centers

on the β5 and ββίεπηβ, which explore how preferences condition the effectiveness of the assigned treatments. If

my arguments above are correct, then I expect to find

a positive and significant β5(indicating greater polariza tion) and a negative and significant β8 (indicating greater moderation).

Table 2 shows strong support for my predictions— the effect of partisan media is conditional on preferences. Like- minded media polarize viewers, but only for subjects who actually want to watch like-minded media. Likewise,

cross-cutting media moderate attitudes only for those

who want to watch these programs. For example, subjects

This content downloaded from 140.211.95.10 on Tue, 13 Oct 2015 05:37:58 UTC All use subject to JSTOR Terms and Conditions

WHY DO PARTISAN MEDIA POLARIZE VIEWERS? 619

who want to watch like-minded media and are assigned to watch them take positions that are 13% more extreme

than those who prefer another type of media (but are as

signed to see like-minded media). Preferences over media

types strongly condition the effectiveness of these exper

imental treatments.16

These findings have important implications for at

least two reasons. First, they show that my results are

not simply the product of randomization bias (Heck man and Smith 1995). One concern in a postbroadcast media environment is that the results in experiment 1

stem simply from showing subjects clips they never would

watch in the real world. But Table 2 illustrates that this is

the exact opposite of what actually happens: it is regular viewers who polarize in response to these programs. This

demonstrates that my findings have a clear real-world

analogue.

Second, and more importantly, they suggest an

important substantive conclusion about the polarizing

power of like-minded media. Those who prefer like

minded media are more politically informed and engaged and also more ex ante attitudinally extreme ( Jamieson and

Cappella 2008; Stroud 2011). This is consistent with the

well-known result that the more informed and engaged

are more extreme (Abramowitz 2010). Thus, like-minded

media take people who are already somewhat extreme and

make them even more extreme. Partisan media increase is

sue polarization not by polarizing moderates, but rather

by increasing polarization among those already away from

the political center. This has particular substantive rele

vance since these more extreme individuals are likely to

make their voices heard in the political process and are

therefore likely to have considerable influence, a point I

return to below. The effects of partisan media are hetero

geneous, and that heterogeneity is quite politically conse

quential.

That said, however, like any moderator-based anal

ysis, these findings also have real, and significant, lim

itations. My results are consistent with the theoretical

argument above about why preferences shape the impact

of the treatment, but they cannot definitively establish

the mechanisms driving these effects—I cannot prove

that these effects occur because of these subjects' differ

ential motivation and cognitive skills. Future work will

be needed to more carefully explore these mechanisms in

more detail, though my findings represent an important initial finding.

16An important topic for future work, however, will be to replicate

experiment 2 using different measures of media preferences and

different designs (such as pre-post designs) to probe the robustness

and limits of these findings.

Table 3 Duration Effects of Partisan Media,

Experiment 3

Time 1 Time 1 Time 2

(All Respondents) (Compliers)

Like-Minded 0.26 0.23 0.24

Treatment (0.10) (0.10) (0.10)

Cross-Cutting 0.08 0.01 0.03

Treatment (0.10) (0.11) (0.10) Constant 0.11 0.14 0.17

(0.07) (0.07) (0.07) N 101 83 78

R-Squared 0.07 0.07 0.09

Note: Cell entries are OLS coefficients with associated standard

errors underneath; see Table 1 for additional details. Results for

time 1 labeled "compilers" show the results only for those who

completed the wave 2 survey.

Are These Just Temporary Artifacts?

Finally, I also investigate the duration of these partisan media effects. Hypothesis 4 argues that because subjects use online processing to update their attitudes, the po larized attitudes resulting from partisan media exposure

should be detectable several days postexposure. I designed

experiment 3 to test this idea. The first wave of experiment 3 follows the standard posttest-only protocol described

earlier. At the close of the experiment, subjects were

asked if they would take part in another survey (wave 2) in exchange for an additional cash payment. In wave 2,

subjects were not shown any additional experimental stimuli and were simply asked to complete the posttest

portion of the initial study again (the wave-1 attitude

items were repeated in wave 2). Wave 2 took place two

days after the initial study, so it can only examine whether

these effects last that long.17 That said, given that the stim

uli are brief news segments, finding any effect even two

days later constitutes important evidence for duration.

Table 3 provides estimates of the immediate effect

of the treatment (measured just after the stimulus was

administered), and the effect in wave 2 a few days later.

Like-minded media have an immediate effect on attitudes

(i.e., there is a positive and significant treatment effect in

wave 1), but consistent with Hypothesis 4, that effect can

still be detected two days later. Indeed, not only is the re

sult statistically significant at time 2, but also the two esti

mates cannot be distinguished from one another (though

17Subjects were sent two requests for participation via email. All

subjects who completed the study did so within two days of the

request, with the majority (83%) doing so on the first day (i.e., two

days after the original study).

This content downloaded from 140.211.95.10 on Tue, 13 Oct 2015 05:37:58 UTC All use subject to JSTOR Terms and Conditions

620 MATTHEWS. LEVENDUSKY

not surprisingly, the effect at time 1 with the entire sam

ple is larger than the effect for compilers only at time 1).

This is quite striking: the effect of watching a short video

stimulus is basically the same two days later as it is the

day subjects watch it.18 Consistent with an account stress

ing online processing, partisan media engage viewers in a

way that has effects postexposure, even several days later.

The effects of partisan media are not simply blips that

fade away immediately after the treatment, but rather are

messages that stay with respondents, at least in the short

term.19

Discussion and Conclusions

This article explores partisan media's effects on attitudi

nal polarization. I argue that viewers' use of motivated

reasoning, combined with the slanted nature of these

programs, generates issue polarization. The results of my

original experiments bear out this theoretical expecta tion. When viewers watch like-minded media that re

inforce their attitudes, they become more extreme, and

these effects persist for at least several days. Further, these

effects are concentrated among the more informed, en

gaged, and extreme segments of the populace who regu

larly watch partisan media programs. Like-minded media

take subjects who are already extreme and make them even

more extreme. Thus, like-minded media polarize not by

making moderate viewers more extreme, but rather by

affecting those already away from the political center.

In contrast, cross-cutting media can either polarize

attitudes—for those with strong attitudes—or moderate

them, for those who prefer to watch cross-cutting me

dia. These different reactions highlight the crucial role a

viewer's abilities and preferences play in how she reacts to

this type of media. My findings suggest that more work is

needed to unpack and explore the heterogeneity in which

viewers choose to watch cross-cutting media, as well as

the consequences of that exposure.

At one level, this sort of media-driven reinforce

ment is nothing new: research on it dates back a half

century to canonical media studies (Berelson, Lazarsfeld, and McPhee 1954; Klapper 1960). The difference with my

18One concern is that subjects might simply remember the items

from time 1 to time 2, making the effects here more about mem

ory than duration. I designed another experiment to rule out this

possibility by only asking the attitude items at time 2. Even here, I

find duration effects (see the appendix).

19Future research should explore how processing style impacts these

effects (Chong and Druckman 2010) as well as how these effects

decay over longer periods of time (Gerber et al. 2011).

findings, however, is that who is doing the activation, and

to what end, is quite distinct. Relatively extreme parti san media hosts are not just priming citizens' underlying

predispositions but are also making them more extreme

and divided. These consequences separate out this type of

partisan media effect from earlier studies of attitude re

inforcement. There is continuity with earlier findings but

also a real difference with this new type of media. Further,

my illustration of the heterogeneity of these findings, and

the political ramifications of these heterogeneous effects, is another novel contribution.

More broadly, my findings suggest how partisan me

dia might contribute to gridlock in American politics. For example, my findings demonstrate how partisan me

dia might help to exacerbate elite polarization. Active and

engaged like-minded viewers are pushed even further to

the extremes on the specific issues discussed on partisan media. If these viewers watch regularly, and are moved

across a host of issues,20 then this can put pressure on

candidates to take more extreme positions on a number

of issues (Layman et al. 2010), which can have real elec

toral and policy consequences, especially given a system of primary elections (Brady, Han, and Pope 2007). For

example, Fox News helped energize Tea Party supporters in 2010, which was a key factor in a number of Republican

primary elections, and has generated policy consequences in Congress (Williamson, Skocpol, and Coggin 2011). Partisan media polarize active and engaged citizens, who

in turn help to fuel elite polarization. Even though most

voters are moderate (Fiorina, Abrams, and Pope 2005) and never hear partisan media messages, by affecting a

more extreme group of individuals, the consequences of

partisan media extend quite broadly.

Of course, elite polarization in and of itself need not

be a bad thing: if there is no clear "best" policy, then

a spirited debate can be a positive development rather than a negative one. But because America's constitutional

system rests on compromise, too much gridlock and de

bate makes governing extremely difficult (Price 2010). Partisan media consumers can put pressure on elected

officials to take extreme positions and eschew bargain

ing with the other side. Some Republican opposition to

President Obama's health care reform legislation was par

tially driven by antireform screeds from partisan media

hosts like Rush Limbaugh—and their listeners—urging

Republican officials to "just say no" (Kurtz 2010; Starr

2010). This kind of message from a key group of sup

porters makes it more difficult for legislators to reach

20This raises the interesting possibility that the issue polarization I find here might spill over into more general kinds of ideological

polarization, though this is a topic for future work.

This content downloaded from 140.211.95.10 on Tue, 13 Oct 2015 05:37:58 UTC All use subject to JSTOR Terms and Conditions

WHY DO PARTISAN MEDIA POLARIZE VIEWERS? 621

across the aisle and strike deals. Indeed, these sorts of

divisions occur not only on significant and divisive is sues like health care reform or increasing the debt ceiling, but also on seemingly prosaic and nonpartisan ones like

flu vaccines (Baum 2011) and evidence-based medicine

(Gerber and Patashnik 2010). Solving problems becomes

less about what is best for the country and more about

what is politically and ideologically expedient (Fiorina

2006). While partisan media alone do not cause these ef

fects, they certainly exacerbate broader trends toward di

vision, gridlock, and dissensus. They accelerate the move

toward the "uncompromising mind" that seeks out grid lock and partisan advantage rather than compromise and

consensus solutions (Gutmann and Thompson 2012). In

this context, the negative consequences of this sort of

media-induced polarization likely outweigh the positive ones.

References

Abramowitz, Alan. 2010. The Disappearing Center. New Haven,

CT: Yale University Press.

Abrams, Dominic, Margaret Wetherell, Sandra Cochrane,

Michael Hogg, and John Turner. 1990. "Knowing What to

Think by Knowing Who You Are." British Journal of Social

Psychology 29(2): 97-119.

Arceneaux, Kevin, and Martin Johnson. 2010. "Does Me

dia Fragmentation Produce Mass Polarization?" Paper pre sented at the annual meeting of the American Political Sci

ence Association, Washington, DC.

Bai, Matt. 2009. "Cable Guise." New York Times, 6 December.

Barker, David. 1999. "Rushed Decisions: Political Talk Ra

dio and Vote Choice, 1994-1996." Journal of Politics 61(2): 527-39.

Baum, Matthew. 2011. "Red State, Blue State, Flu State: Media

Self-Selection and Partisan Gaps in Swine Flu Vaccinations."

Journal of Health Politics, Policy, and Law 36(6): 1021-59.

Baum, Matthew, and Tim Groeling. 2008. "New Media and

the Polarization of American Political Discourse." Political

Communication 25(4): 345-65.

Baum, Matthew, and Tim Groeling. 2010. War Stories. Prince

ton, Ν J: Princeton University Press.

Bennett, W. Lance, and Shanto Iyengar. 2008. "A New Era of

Minimal Effects?" Journal of Communication 58(4): 707-31.

Berelson, Bernard, Paul Lazarsfeld, and William McPhee. 1954.

Voting. Chicago: University of Chicago Press.

Brady, David, Hahrie Han, and Jeremy Pope. 2007. "Primary Elections and Candidate Ideology: Out of Step with the

Primary Electorate?" Legislative Studies Quarterly 32(1):

79-105.

Campbell, Angus, Philip Converse, Warren Miller, and Donald

Stokes. [ 1960] 1980. The American Voter. Chicago: University of Chicago Press, Midway Reprint.

Chong, Dennis, and James Druckman. 2010. "Dynamic Public

Opinion: Framing Effects over Time." American Political

Science Review 104(4): 663-80.

Dalton, Russell, Paul Beck, and Robert Huckfeldt. 1998. "Parti

san Cues and the Media: Information Flows in the 1992 Pres

idential Election." American Political Science Review 92(1): 111-26.

DellaVigna, Stefano, and Ethan Kaplan. 2007. "The Fox News

Effect: Media Bias and Voting." The Quarterly Journal of Economics 122(3): 1187-1234.

Ditto, Peter, and David Lopez. 1992. "Motivated Skepticism: Use of Differential Decision Criteria for Preferred and Non

preferred Conclusions." Journal of Personality and Social Psy

chology 63(4): 568-84.

Druckman, James, Jordan Fein, and Thomas Leeper. 2012.

"A Source of Public Opinion Stability." American Political

Science Review 106(2): 430-54.

Druckman, James, and Kjersten Nelson. 2003. "Framing and Deliberation: How Citizens' Conversations Limit Elite

Influence." American Journal of Political Science 47(4): 729-45.

Fiorina, Morris. 2006. "Parties as Problem Solvers." In Promot

ing the General Welfare, ed. Alan Gerber and Eric Patashnik.

Washington, DC: Brookings Press, 237-55.

Fiorina, Morris, Samuel Abrams, and Jeremy Pope. 2005. Cul

ture War? The Myth of a Polarized America. New York: Person

Longman.

Gaines, Brian, and James Kuklinski. 2011. "Experimental Es

timation of Heterogeneous Treatment Effects Related to

Self-Selection." American Journal of Political Science 55(3): 724-36.

Gentzkow, Matthew, and Jesse Shapiro. 2010. "What Drives

Media Slant? Evidence from U.S. Newspapers." Econometrica

78(1): 35-71.

Gerber, Alan, James Gimpel, Donald Green, and Daron Shaw.

2011. "How Large and Long-Lasting Are the Persuasive Ef

fects of Televised Campaign Ads?" American Political Science

Review 105(1): 135-50.

Gerber, Alan, and Eric Patashnik. 2010. "Problem Solving in

a Polarized Age: Comparative Effectiveness Research and

Politicization of Evidence-Based Medicine." The Forum8( 1 ).

Goren, Paul, Christopher Federico, and Miki Caul Kitti. 2009.

"Source Cues, Partisan Identities, and Political Value Expres sion." American Journal of Political Science 53(4): 805-20.

Groseclose, Timothy, and Jeffrey Milyo. 2005. "A Measure

of Media Bias." Quarterly Journal of Economics 120(4): 1191-1237.

Gutmann, Amy, and Dennis Thompson. 2012. The Spirit of

Compromise. Princeton, Ν J: Princeton University Press.

Heckman, James, and Jeffrey Smith. 1995. "Assessing the Case

for Social Experiments." Journal of Economic Perspectives

9(2): 85-110.

Holbert, R. Lance, R. Kelly Garrett, and Laurel Gleason. 2010.

"A New Era of Minimal Effects? A Response to Bennett and

Iyengar." Journal of Communication 60(1): 15-34.

Iyengar, Shanto, and Kyu Hahn. 2009. "Red Media, Blue Media:

Evidence of Ideological Selectivity in Media Use." Journal of Communication 59(1): 19-39.

This content downloaded from 140.211.95.10 on Tue, 13 Oct 2015 05:37:58 UTC All use subject to JSTOR Terms and Conditions

622 MATTHEW S. LEVENDUSKY

Jamieson, Kathleen Hall, and Joseph Cappella. 2008. Echo

Chamber: Rush Limbaugh and the Conservative Media Es

tablishment. New York: Oxford University Press.

Jamieson, Kathleen Hall, Bruce Hardy, and Daniel Romer. 2007.

"The Effectiveness of the Press in Serving the Needs of

American Democracy." In A Republic Divided: The Annen

berg Democracy Project, ed. Kathleen Hall Jamieson. New

York: Oxford University Press, 21-51.

Keith, Bruce, David Magleby, Candice Nelson, Elizabeth Orr,

Mark Westlye, and Raymond Wolfinger. 1992. The Myth of the Independent Voter. Berkeley and Los Angeles: University of California Press.

Klapper, Joseph. 1960. The Effects of Mass Communication.

Glencoe, IL: Free Press.

Klayman, Joseph, and Young-Won Ha. 1987. "Confirmation,

Disconfirmation, and Information in Hypothesis Testing."

Psychological Review 94(2): 211-228.

Krosnick, Jon, and Richard Petty. 1995. Attitude Strength: An

tecedents and Consequences. Hillsdale, NJ: Erlbaum.

Kunda, Ziva. 1990. "The Case for Motivated Reasoning." Psy

chological Bulletin 108(3): 480-98.

Kurtz, Howard. 2010. "Conservative David Frum Loses Think

Tank Job After Criticizing GOP." Washington Post, 26 March.

Ladd, Jonathan, and Gabriel Lenz. 2009. "Exploiting a Rare

Communication Shift to Document the Persuasive Power of

the News Media." American Journal of Political Science 53(2): 394-410.

Layman, Geoffrey, Thomas Carsey, John Green, Richard

Herrera, and Rosalyn Cooperman. 2010. "Activists and Con

flict Extension in American Party Politics." American Politi

cal Science Review 104(2): 324-46.

Lee, Eun-Ju. 2007. "Deindividualization Effects on Group Po

larization in Computer-Mediated Communication" Journal

of Communication 57(2): 385-403.

Lodge, Milton, Kathleen McGraw, and Patrick Stroh. 1991. "An

Impression-Driven Model of Candidate Evaluation." Amer

ican Political Science Review 83(2): 399-419.

Lodge, Milton, Marco Steenbergen, and Shawn Brau. 1995. "The

Responsive Voter: Campaign Information and the Dynamics of Candidate Evaluation." American Political Science Review

89(2): 309-26.

Lodge, Milton, and Charles Taber. 2001. "Three Steps Toward

a Theory of Motivated Reasoning." In Elements of Reason,

ed. Arthur Lupia, Mathew McCubbins, and Samuel Popkin. New York: Cambridge University Press, 183-213.

Lodge, Milton, and Charles Taber. 2005. "The Automaticity of Affect for Political Leaders, Groups, and Issues: An Ex

perimental Test of the Hot Cognition Hypothesis." Political

Psychology 26(3): 455-82.

Lupia, Arthur, and Mathew McCubbins. 1998. The Democratic

Dilemma: Can Citizens Learn What They Need to Know? New

York: Cambridge University Press.

Morris, James, Nancy Squires, Charles Taber, and Milton Lodge. 2003. "Activation of Political Attitudes: A Psychophysiologi cal Examination of the Hot Cognition Hypothesis." Political

Psychology 24(4): 727—45.

Moscovici, Serge, and Marisa Zavalloni. 1969. "The Group as

Polarizer of Attitudes." Journal of Personality and Social Psy

chology 12(2): 125-35.

Mullainathan, Sendhil, and Andrei Shleifer. 2005. "The Market

for News." American Economic Review 95(4): 1031-53.

Mutz, Diana. 2006. "How the Mass Media Divide Us." In

Red and Blue Nation? Volume 1, ed. David Brady and

Pietro Nivola. Stanford, CA: Hoover Institution Press, 223-48.

Mutz, Diana. 2007. "Effects of Ίη-Your-Face' Television Dis

course on Perceptions of a Legitimate Opposition." Ameri

can Political Science Review 101(4): 621-35.

Pew Center for the People and the Press. 2010. "Americans

Spending More Time Following the News." 12 September,

http: //people-press.org/report/652/.

Pomerantz, Eva, Shelly Chaiken, and Rosalind Tordesillas. 1995.

"Attitude Strength and Resistance Processes." Journal of Per

sonality and Social Psychology 69(3): 408-19.

Price, David. 2010. "After the 'Housequake': Leadership and

Partisanship in the Post-2006 House." The Forum 8(1), article 6.

Price, Vincent. 1989. "Social Identification and Public Opinion: Effects of Communicating Group Conflict." Public Opinion

Quarterly 53(2): 197-224.

Prior, Markus. 2007. Post-Broadcast Democracy: How Me

dia Choice Increases Inequality in Political Involvement

and Polarizes Elections. New York: Cambridge University Press.

Project for Excellence in Journalism. 2009. "The State of the

News Media: An Annual Report on American Journalism."

http://www.stateofthemedia.org.

Redlawsk, David. 2001. "You Must Remember This: A Test

of the Online Model of Voting." Journal of Politics 63(1): 29-58.

Redlawsk, David. 2002. "Hot Cognition or Cool Considera

tion: Testing the Effects of Motivated Reasoning on Po

litical Decision Making." Journal of Politics 64(4): 1021—

44.

Rosensteil, Thomas. 2006. "How the Mass Media Divides Us—

Two Alternative Perspectives." In Red and Blue Nation? Vol

ume 1, ed. Pietro Nivola and David Brady. Stanford, CA:

Hoover Institution Press, 249-54.

Starr, Paul. 2010. "Governing in the Age of Fox News." The

Atlantic, January/February, 95-98.

Steinberg, Jacques. 2007. "Cable Channel Nods to Ratings and

Leans Left." New York Times, 6 November.

Stroud, Natalia. 2010. "Polarization and Partisan Selective Ex

posure." Journal of Communication 60(3): 556-76.

Stroud, Natalia. 2011. Niche News. New York: Oxford University Press.

Sunstein, Cass. 2009. Going to Extremes. New York: Oxford

University Press.

Taber, Charles, Damon Cann, and Simona Kucsova. 2009. "The

Motivated Processing of Political Arguments." Political Be

havior 31(2): 137-55.

Taber, Charles, and Milton Lodge. 2006. "Motivated Skepticism in the Evaluation of Political Beliefs." American Journal of Political Science 50(3): 755-69.

Torgerson, David, and Bonnie Sibbald. 1998. "Understanding Controlled Trials: What Is a Patient Preference Trial?" BMJ

316(360): 360.

This content downloaded from 140.211.95.10 on Tue, 13 Oct 2015 05:37:58 UTC All use subject to JSTOR Terms and Conditions

WHY DO PARTISAN MEDIA POLARIZE VIEWERS? 623

Williamson, Vanessa, Theda Skocpol, and John Coggin. 2011.

"The Tea Party and the Remaking of Republican Conser

vatism." Perspectives on Politics 9(1): 25-43.

Zaller, John. 1992. The Nature and Origins of Mass Opinion. New York: Cambridge University Press.

Zaller, John. 1996. "The Myth of Massive Media Impact Re

vived." In Political Persuasion and Attitude Change, ed. Di

ana Mutz, Paul Sniderman, and Richard Brody. Ann Arbor:

University of Michigan Press, 17-79.

Supporting Information

Additional Supporting Information may be found in the

online version of this article at the publisher's web site:

Online Appendix: "Why Do Partisan Media Polar

ize Voters?"

• Table Al: Descriptive Statistics, Dependent Vari

able, by Experiment

• Table A2: Testing For Differential Treatment Ef

fects by Party, Experiment 1 • Table A3: Fully Saturated Model, with Indepen

dents

• Table A4: Validating the Measure of Media Prefer

ences

• Table A5: Partisan Media, Attitude Extremity, and

Participation • Table A6: Partisan Media Consumption and Polit

ical Information, Pew Data • Table A7: Political/Campaign Interest, NAES 2008

Data

• Table A8: Effect of Partisan Media on Attitude Ex

tremity, 2008 NAES Data • Table A9: Effects of Partisan Media Relative to an

Apolitical Control

• Table A10: Editorial Experiment Results

• Table Al 1: Attitude Durability over Three Days

This content downloaded from 140.211.95.10 on Tue, 13 Oct 2015 05:37:58 UTC All use subject to JSTOR Terms and Conditions