Media and Political Polarization

This article examines if the emergence of more partisan media has contributed to political polarization and led Americans to support more partisan policies and candidates. Congress and some newer media outlets have added more partisan messages to a continuing supply of mostly centrist news. Although political attitudes of most Americans have remained fairly moderate, evidence points to some polarization among the politically involved. Proliferation of media choices lowered the share of less interested, less partisan voters and thereby made elections more partisan. But evidence for a causal link between more partisan messages and changing attitudes or behaviors is mixed at best. Measurement problems hold back research on partisan selective exposure and its consequences. Ideologically one-sided news exposure may be largely confined to a small, but highly involved and influential, segment of the population. There is no firm evidence that partisan media are making ordinary Americans more partisan. 101 A nn u. R ev . P ol it. S ci . 2 01 3. 16 :1 01 -1 27 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g by P ri nc et on U ni ve rs ity L ib ra ry o n 05 /1 4/ 13 . F or p er so na l u se o nl y. PL16CH06-Prior ARI 7 April 2013 1:28 MEDIA AND POLITICAL POLARIZATION Has the emergence of more partisan media created political polarization and led the American public to support more partisan policies and candidates? Concern is growing about the pernicious impact of fervently populist or ideological rhetoric displayed on cable news, talk radio, and the Internet. Pundits, commentators, and some social scientists worry about the influence of firebrands like Rush Limbaugh, Sean Hannity, and Keith Olbermann. Some fear that less knowledgeable Americans are particularly seduced by the seeming simplicity of their populist worldviews. Others suspect that ideologically unambiguous content increasingly attracts viewers and listeners who share the hosts’ political leanings, thus reinforcing partisan views and contributing to political polarization. These are important concerns, especially in a heated political climate in which the fringes appear eager to escalate. By somemeasures, the general public appearsmorepartisan aswell. Causally linking partisan messages from media and political elites with growing mass polarization is tempting. Yet, there are also good theoretical reasons to expect that blatantly partisan messages will leave public opinion mostly unchanged—because citizens ignore them, resist them, or take them for granted. Empirical research on the influence of elite andmediamessages onmass political behavior faces formidable challenges,many related to the very technological changes that brought us BillO’Reilly and The Huffington Post. Political scientists, communication scholars, and economists have begun to examine persuasion and selective exposure in the current high-choice media environment. This review summarizes existing evidence and concludes that empirical analysis is severely hampered by a seemingly simple problem: we do not know how many and what kind of people are exposed to which messages. The starting point for this account, documented in the next section, is evidence that Congress and some newer media outlets add more partisan messages to the continuing supply of mostly centrist news produced by many large outlets. Next, a review of public opinion research finds that the political attitudes of most Americans have remained fairly centrist, but signs of polarization emerge among the most politically involved. The remainder of this article evaluates evidence concerning the causal impact of a more partisan media environment on mass polarization. Compositional changes in the voting public link cable penetration and the polarization of elections even before the founding of Fox News and the creation of the World Wide Web. Yet, there is little evidence that more partisanmessages changed people’s attitudes or behaviors. A key concept, discussed in detail here, is selective exposure. Evidence for partisan selective exposure is mixed and does not, on its own, show media impact. In this area in particular, measurement problems loom large, making it difficult to quantify selectivity. As it is surprisingly difficult to pinpoint how many people follow partisan news and for how long, a section of this article describes the cable news audience in some detail. The last section examines new research designs that consider the impact of partisan media on attitudes and voting behavior. The conclusion discusses the growing importance of news brands and social media, and ends by pointing out the need for more research on political activists and their use of partisan media. PARTISAN ELITES, PARTISAN MEDIA? It is easy to see why the claim of growing media influence resonates. Political elites have become more polarized, and some new voices on cable television and on the web offer ideologically slanted content. Among the less controversial propositions in this research area is the claim that partisan messages have become more common in the past quarter century. There is compelling evidence that the two parties in Congress oppose each other more often and more consistently. The percentage of roll-call votes in which a majority of one party opposes a majority of the other 102 Prior A nn u. R ev . P ol it. S ci . 2 01 3. 16 :1 01 -1 27 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g by P ri nc et on U ni ve rs ity L ib ra ry o n 05 /1 4/ 13 . F or p er so na l u se o nl y. PL16CH06-Prior ARI 7 April 2013 1:28 party has risen (e.g., Rohde 1991, Sinclair 2000). Interest groups rate members of Congress as more ideologically split (e.g., Fleisher & Bond 2000). To the extent that the underlying dimension of competition persistently reflects ideology, roll-call votes show growing ideological divergence. Whereas several Republican members of Congress were ideologically more liberal than the most conservative Democrats two or three decades ago, all Republicans are to the right of all Democrats today (McCarty et al. 2006, Poole & Rosenthal 1997). As formedia content, there is no evidence that longstanding outlets have becomemore partisan. Evening newscasts on the broadcast networks, long the most widely followed news source, are mostly centrist with possibly aminor tilt in the liberal direction. Ameta-analysis of about twodozen content analyses finds a small but fairly consistent Democratic advantage on amount and tone of presidential campaign coverage between 1968 and 1996 (D’Alessio & Allen 2000).1 Hamilton (2004) shows that network news reflects the relatively liberal priorities of younger women because they are most valued by advertisers and most indifferent about watching network news. Groeling (2008) examines bias by comparing the circumstances in which newscasts reported presidential approval ratings for Bill Clinton and George W. Bush measured by their own polls. Broadcast networks were more likely to report a poll when it showed declining approval for President Bush or increasing approval for President Clinton. Few content analyses of CNN exist for the period when it was the only cable news network. One finds CNN campaign coverage to bemostly centrist in 1992 and 1996 (Lowry & Shidler 1998). Groseclose & Milyo (2005) propose a way to score media outlets and members of Congress on the same ideological scale. They use the relative frequency of references to different think tanks in news coverage and the Congressional Record to link outlets and legislators. The more a news outlet refers to think tanks that are predominantly cited by liberal legislators, the more liberal the news outlet is inferred to be. The advantage of Groseclose & Milyo’s method is that researchers do not have to make subjective judgments about the slant of media coverage. The disadvantage is that reliance on citation patterns makes assumptions that are poorly understood (how do legislators and journalists pick which think tank to cite?) and not easy to implement (how does one eliminate citations used only to criticize a think tank?). Gentzkow & Shapiro (2010) employ a variation of the same approach that uses short word combinations as bridges between newspapers and members of Congress, but they find a notably different slant for the Wall Street Journal. According to Groseclose &Milyo (2005), the news reporting of theWall Street Journal is more liberal than even theNew York Times. InGentzkow&Shapiro’s (2010) estimates, in contrast, the Wall Street Journal is more conservative than the New York Times and almost as conservative as theWashington Times. Gasper (2011) shows that Groseclose &Milyo’s estimates are not robust to the deletion of one prominent think tank and that estimating slant over time generates strong trends that are difficult to explain. These inconsistencies cast some doubt onGroseclose &Milyo’s (2005, p. 1,192) verdict of “strong liberal bias.” The more important empirical finding that emerges consistently in these studies is that most large media outlets are centrist compared to members of Congress. In 1999, only 3 out of 100 senators had ideological (Americans for Democratic Action) scores between 33 and 67 on a 0– 100 scale (Groseclose & Milyo 2005, p. 1,216). Yet 16 of 20 media outlets fall within this range according to Groseclose & Milyo and 17 of 20 according to Gasper. Eighteen of 20 outlets have ideological positions between JoeLieberman andSusanCollins, who are among themostmoderate 1Not included in the meta-analysis is a recent study that finds evidence for Republican bias in visual presentation style of network coverage of presidential elections from 1992 to 2004 (Grabe & Bucy 2009). For example, Republican candidates received about the same amount of visual coverage as Democratic candidates but were shown from low camera angles more frequently, which has been demonstrated to increase candidate evaluations. www.annualreviews.org • Media and Political Polarization 103 A nn u. R ev . P ol it. S ci . 2 01 3. 16 :1 01 -1 27 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g by P ri nc et on U ni ve rs ity L ib ra ry o n 05 /1 4/ 13 . F or p er so na l u se o nl y. PL16CH06-Prior ARI 7 April 2013 1:28 senators (Groseclose & Milyo 2005, p. 1,228). One reason for centrism is that newspapers draw on many different reporters, thus introducing variation in slant within papers (Dalton et al. 1998, Shaw & Sparrow 1999). Groseclose &Milyo (2005), Gasper (2011), and Gentzkow & Shapiro (2010) all examine news reporting and exclude editorials. Ho & Quinn (2008) analyze newspaper editorials on Supreme Court decisions to compare ideology of newspapers and justices on the same scale. They find only modestly less centrism. During the Rehnquist court, about half of all newspapers were located between Justices Kennedy and Breyer, the secondand third-most centrist justices at the time. Some papers take strongly ideological positions, however. TheNew York Times’ editorial stance is estimated to be more liberal than the most liberal justice, and theNew York Post, Investor’s Business Daily, and theWashington Times are almost as conservative as Justice Scalia, the second-most conservative justice. Again, however, bundling generates moderation. In their content analysis of 41 newspapers during the 1992 presidential campaign, Dalton et al. (1998) find very low correlations between tone of editorials and news reporting. It is telling that the four most ideological papers according to Ho & Quinn (2008) either circulate nationally or in markets with multiple big papers. Most newspapers in the United States publish in one-paper markets and thus cater to an ideologically heterogeneous audience (see Gentzkow & Shapiro 2010, Hamilton 2004, Petrova 2011). Cable and Internet outlets that entered a crowded national market more recently have weaker economic incentives to aim for a politically moderate median user. Groeling (2008) finds that Special Report on the Fox News Channel (FNC) was disproportionately likely to report polls that showed declines in approval of President Clinton. Baum&Groeling (2008) show that http://www.DailyKos.com is muchmore likely to pick up wire stories favorable to Democrats, whereas http://www.FoxNews.com and http://www.FreeRepublic.com often avoid these stories and present pro-Republican content. When political blogs link to other blogs, they mostly pick ideologically congenial ones (Adamic & Glance 2005, Hargittai et al. 2008). Analyzing dozens of the most popular opinion formats on talk radio, cable, and blogs over a 10-week period in 2009, Sobieraj & Berry (2011) find a heavy dose of insulting language, name calling, “very dramatic negative exaggeration,” and mockery. Cable and talk radio shows contain dozens of such instances of “outrage” per hour (for qualitative illustrations, see Jamieson & Cappella 2008). These smaller, more specialized, opinion-focused new media outlets provide the greatest opportunities for one-sided media exposure. In summary, most large US media outlets are politically centrist and provide a balance of competing viewpoints. But the first condition for growing mass polarization through increasingly partisan media is partially met: some talk radio shows, cable news channels, and websites do offer more ideologically extreme packages of news and opinion. The next section examines the second condition: has the mass public become more politically polarized since more partisan media began to emerge? POLARIZATION AND PARTISANSHIP IN THE MASS PUBLIC In reviewing evidence for polarization and greater partisanship in the American public, it is important to distinguish between attitudes, perceptions, and behaviors. Voting in a more partisan way does not necessarily indicate stronger partisan attitudes if choices no longer include moderate options. Perceptions of rising polarization may reflect such evolving choices, without telling us much about the perceiver’s partisanship. Evidence for attitude polarization—individuals changing their issue positions, ideological convictions, or partisan sentiments to produce less centrist, more sharply opposed aggregate distributions of the most politically relevant attitudes—turns out to be ambiguous. 104 Prior A nn u. R ev . P ol it. S ci . 2 01 3. 16 :1 01 -1 27 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g by P ri nc et on U ni ve rs ity L ib ra ry o n 05 /1 4/ 13 . F or p er so na l u se o nl y. PL16CH06-Prior ARI 7 April 2013 1:28 Abundant evidence has emerged that Americans’ vote choices are more strongly related to partisan considerations. Bartels (2000) shows that party identification (ID) has become a better predictor of vote decisions since the mid-1970s, and voters today are less likely to split their ticket (Hetherington 2001, Mayer 1998). In the aggregate, stronger partisanship is seen in the resurging correlation between the district-level presidential vote and the vote for the House candidate of the president’s party (Fleisher & Bond 2004; Jacobson 2000, 2003). As a result of these trends, the number of districts that elect a representative from one party but give amajority to the presidential candidate of the opposite party has declined ( Jacobson 2003). The declining volatility of election outcomes since the 1970s (Bartels 1998, pp. 295–97) is also consistent with stronger partisan influence on voting decisions. Fiorina (2002, Fiorina & Abrams 2008) has noted that these empirical patterns could be caused by starker, more ideologically coherent parties and candidates alone. Democratic and Republican elites have adopted positions on cultural issues that line up better with their established standpoints on economic issues (Fiorina 2006, Layman & Carsey 2002a). As Republican and Democratic alternatives have diverged from each other ideologically to offer voters a starker choice, an increase in the impact of party ID could emerge even in the absence of changes among voters. It does not necessarily mean individuals have become more partisan or assign greater importance to voting in accordance with their party ID. More Americans do in fact report that they see important differences between the parties (Hetherington 2001; Prior 2007, p. 222). And on many political issues, more respondents are able to place the Republican Party to the right of the Democratic Party than in previous decades (Abramowitz & Saunders 1998, Layman&Carsey 2002a, Levendusky 2009b). Overall evaluations of parties have diverged (e.g., Prior 2007, p. 222), and party identifiers rate the opposing party increasingly negatively while not raising evaluations of their own party (Iyengar et al. 2012). Rising out-group dislike may be expected as a reaction to elite divergence, but no causal link has been demonstrated. The notion that more partisan voting behavior is largely a result of a more partisan choice set, not more partisan voters, receives support from the negative findings on attitude polarization. DiMaggio et al. (1996) define four dimensions of polarization in the mass public: the dispersion of attitudes, the extent to which attitudes cluster around two contrasting positions with fewmoderate views in between, the link between different issue positions (“ideological polarization”), and the existence of systematic differences between subpopulations (“identity-based polarization”). They find mostly convergence, not polarization, on the first three dimensions. Many opinion distributions have become less dispersed and more centered on one middle position (see also Evans 2003). On most issues, Americans take moderate positions, and their disagreement has not intensified noticeably in recent decades (Fiorina 2006). Opinions of different demographic groups have in fact grown more similar to an extent that leads DiMaggio et al. (1996, p. 738) to proclaim “dramatic depolarization in intergroup differences.” Increases in issue constraint (Converse 1964)—taking consistently liberal or conservative positions on sets of issues—have been minimal (Baldassarri & Gelman 2008). In the electorate as a whole, people’s issue attitudes continue to be characterized better by three different dimensions than one left-right continuum (Layman & Carsey 2002b). The belief systems of some Americans do appear to have changed in one way, however. Democrats are increasingly likely to take liberal positions on salient issues and call themselves liberal; Republicans, likewise, have become more conservative (e.g., Abramowitz & Saunders 1998, Layman&Carsey 2002a, Baldassarri &Gelman 2008, Stoker & Jennings 2008, Levendusky 2009b). A considerable amount of this supposed change is in fact measurement error (Carsey & Layman 2006, Levendusky 2009a), but some people do exhibit stronger links than in the past www.annualreviews.org • Media and Political Polarization 105 A nn u. R ev . P ol it. S ci . 2 01 3. 16 :1 01 -1 27 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g by P ri nc et on U ni ve rs ity L ib ra ry o n 05 /1 4/ 13 . F or p er so na l u se o nl y. PL16CH06-Prior ARI 7 April 2013 1:28 between issue positions and their party ID or ideology. This trend could emerge because people change their issue positions to make them consistent with their party ID or because people change their party ID to accord with their issue positions. If Republicans (or conservatives) and Democrats (or liberals) change at roughly the same rate, both processes leave the distribution of issue positions in the population unchanged but widen issue differences between Republicans and Democrats—the only kind of “identity-based polarization” that DiMaggio et al. (1996) observe. Of the two possible processes that generate a closer match of party ID and issue positions, changes in issue positions—what Layman & Carsey (2002b) call “party-based issue conversion” and Levendusky (2009b) refers to as “party-driven sorting”—appear to be more common (Levendusky 2009a,b), especially when the issue is not particularly important to the individual (Carsey & Layman 2006). This finding squares with the strong accumulated evidence that party ID is highly stable (e.g., Converse & Markus 1979, Green et al. 2002) and powerfully organizes many other components of people’s belief systems, including their core values (Goren 2005) and beliefs about objective conditions such as the crime rate or the state of the economy (Bartels 2002). There are two notable exceptions to the prevalence of party-based sorting. In response to civil rights legislation, many Southern conservatives became dissatisfied with the increasingly liberal positions of the Democratic Party and became Republicans (Green et al. 2002, Levendusky 2009b). Second, on the issue of abortion, people are more likely than on other issues to resolve inconsistencies by modifying their party identification (Levendusky 2009b). Abortion is also the only issue on which DiMaggio et al. (1996, Evans 2003) found clear evidence of polarization, and even Fiorina (2006, pp. 88–90) shows sizable party differences. There is stronger evidence for attitude polarization among the most politically engaged, most partisan Americans. Among partisans, the correlation between the three main dimensions of issue attitudes (economic, cultural, racial) appears to have increased recently (Layman & Carsey 2002a,b). Layman & Carsey (2002b) use panel data to show that adjustments of party ID or issue positions occur only among those respondents who perceive elite-level issue differences. Other scholars have also found disproportionate polarization among more engaged segments of the electorate. The correlation between party ID and ideology is considerably higher for more politically knowledgeable individuals (Abramowitz & Saunders 2008, pp. 545–48; Hetherington 2009, p. 438). Evans (2003, pp. 80–81) finds that “the politically active are becoming more polarized—and particularly polarized on the most political of matters, feelings towards liberals and conservatives.” Baldassarri & Gelman (2008) find some evidence for increasing belief consistency among politically interested Americans. Jacobson (2000, pp. 22–23) shows that the ideological distance between Republican and Democratic “activists” (the roughly 20% of House voters who report engaging in multiple political activities such as working for a candidate or donating money) has increased considerably more than the distance between partisan House voters who are not as engaged in the campaign. It is critically important to distinguish more partisan vote choice from more partisan political attitudes. Asmentioned above, partisan elites offer amore ideologically coherent choice today than several decades ago, and there is clear evidence that people vote for their side more reliably, but there is little compelling evidence that this more partisan voting reflects more partisan attitudes, at least for the large majority of Americans. Having more ideologically coherent parties to choose from does not make you more partisan, just as buying tofu when the store is out of meat does not make you a vegetarian. Only among the politically engaged is evidence for attitude polarization somewhat stronger. The second condition for growing mass polarization through increasingly partisan media is thus met only for a minority of Americans, albeit an influential minority. But did 106 Prior A nn u. R ev . P ol it. S ci . 2 01 3. 16 :1 01 -1 27 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g by P ri nc et on U ni ve rs ity L ib ra ry o n 05 /1 4/ 13 . F or p er so na l u se o nl y. PL16CH06-Prior ARI 7 April 2013 1:28 partisan media have anything to do with the sorting of engaged citizens into more ideologically coherent blocs or the more partisan voting in the larger electorate? POLARIZATION WITHOUT PERSUASION: MEDIA CHOICE AND POLITICAL INVOLVEMENT An important piece of empirical evidence complicates accounts in which mass polarization results from more ideologically coherent partisan elites. One component of mass polarization, partisan voting, increased as a function of cable penetration rates. A rise in the share of households with cable television in a media market was followed by an increase in partisan voting, as measured by the correspondence between presidential and House voting patterns in the media market or by the correlation between consecutive presidential election results (Prior 2007, pp. 239–44). If more partisan electoral behavior is caused by parties or candidates taking positions that are more consistently conservative or liberal, why does it occur at greater rates in areaswith high cable penetration? The relationship between cable access and partisan voting also poses a challenge for explanations that see more ideological media content as the cause of polarization. Cable access increased the impact of partisanship in elections, but this relationship was first observed in the 1970s, when cable systems did not carry any news channels. Its impact strengthened in the 1980s, the decade with the steepest growth in cable penetration. CNN began operating in 1980, but the emergence of Fox News and MSNBC was still more than a decade away. How did cable television polarize elections before the advent of slanted news? Neither elite polarization nor media persuasion is necessary to explain how changes in the media environment can generate more partisan voting behavior. The expansion of media choice can polarize elections in the absence of any attitude change, reinforcement, and polarization. In the 1960s and 1970s, the glory days of broadcast television, more people routinely watched television news than in any other period. Compared to print media, broadcast television helped less educated viewers learn more about politics. Even people with little interest in news and politics watched network newscasts because they were glued to the set and there were no real alternatives to news in many markets during the dinner hour. News exposure motivated some of these less educated, less interested viewers to go to the polls. And because their political views were not particularly ideological or partisan, their votes reduced the aggregate impact of party ID, so elections were less partisan in the broadcast era (Prior 2007, pp. 37–91, 163– 213). Beginning in the 1970s, the growth of cable television and later the Internet created an escape for this inadvertent and nonideological audience segment: more entertainment options that were now easily available when broadcast networks offered newscasts. With greater media choice, individual content preference became increasingly important in determining who would watch the news and who would abandon it. Without their inadvertent news exposure, entertainment fans lacked the occasional push to the polls. The turnout gap between news and entertainment fans widened. And because entertainment fans are less partisan, their dropping turnout rates led to more partisan elections. The stronger partisan preferences of remaining voters reduced the volatility of election outcomes and raised the aggregate impact of partisanship. Greater media choice polarized elections even before choices began to include more partisan news and opinion formats (Prior 2007, pp. 94–138, 214–48). If the goal was to find a connection between media and more partisan elections, we can stop looking. The culprit turns out to be not Fox News, but ESPN, HBO, and other early cable channels that lured moderates away from the news—and away from the polls. Polarization without persuasion—through technology-induced compositional change of the voting public and www.annualreviews.org • Media and Political Polarization 107 A nn u. R ev . P ol it. S ci . 2 01 3. 16 :1 01 -1 27 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g by P ri nc et on U ni ve rs ity L ib ra ry o n 05 /1 4/ 13 . F or p er so na l u se o nl y. PL16CH06-Prior ARI 7 April 2013 1:28 elite-induced clarification of electoral choices—is sufficient to explain why elections have become more partisan and moderates have all but disappeared in Congress. This explanation has the advantage of not requiring attitude polarization, for which evidence is weak, or media persuasion, which does not happen easily. But is it the whole story? Even if partisan media are not necessary to explain why elections have become more partisan, they may have exacerbated the trend or polarized attitudes. DO PARTISAN MEDIA INCREASE ATTITUDE POLARIZATION? Theories in political and cognitive psychology suggest severalmechanisms to explainwhy exposure to partisan media might, or might not, lead to attitude polarization and more partisan voting behavior. One mechanism concerns the impact of exposure on attitudes: when a conservative, a moderate, and a liberalwatch a conservativemedia outlet, what happens to their attitudes?A second mechanism covers the exposure stage: under what circumstances would a moderate or a liberal watch a conservative outlet in the first place? The first topic is often referred to as persuasion, the second as selective exposure. A voluminous literature has accumulated on both topics over more than half a century. This review focuses on studies that explicitly cover partisan messages. Persuasion: The Effect of Exposure on Attitudes InZaller’s (1992, 1996) influentialReceive-Accept-Sample (RAS)model of elite influence,whether a counter-attitudinal message changes a recipient’s attitude depends on the recipient’s political sophistication and her knowledge of the message source. Politically sophisticated recipients are less likely to be swayed by a counter-attitudinal message because they have a greater store of information that might contradict, and thus neutralize, the message. Recipients with knowledge of the message source are less likely to be swayed by a counter-attitudinal message because they question the credibility of the information. A moderate or liberal exposed to a conservative message could change her attitude in a conservative direction (if she holds no information to counter-argue the message and does not know the message source) or remain unmoved (if she counter-argues or dismisses the message based on source cues). Theories of motivated reasoning (Lodge & Taber 2000, Redlawsk 2002, Taber & Lodge 2006, Druckman et al. 2012), which emphasize affective and automatic processes, lead to similar predictions. Unlike the RAS model, motivated reasoning explicitly allows attitudes to change in the direction opposite the message, i.e., a conservative message can make a liberal’s attitude even more liberal. In the RAS model, this outcome occurs only stochastically through the “sampling” process. For a conservative exposed to a conservative message, Zaller’s RAS model predicts attitude moderation if the recipient’s initial attitude is more conservative than the message. In all other scenarios, attitudes of conservative recipients become more conservative or remain unchanged (if the recipient holds information that contradicts themessage or if the recipient is alreadymaximally conservative). Attitudepolarization in response to partisanmedia exposure is thus a typical outcome in Zaller’s RASmodel and theories of motivated reasoning. Recent work onmedia effects provides empirical examples (e.g., Arceneaux & Johnson 2010, Levendusky 2012b). Outcomes other than attitude polarization as a result of partisan news exposure are also theoretically plausible in the RAS model, however. Persuasion—acceptance of counter-attitudinal messages—arises in the RAS model when message recipients lack prior knowledge to counterargue the message and do not recognize its source. In the motivated reasoning framework, Taber & Lodge (2006, p. 756) caution that attitude polarization may not occur when a message fails to “arouse sufficient partisan motivation to induce much biased processing.” Instead of polarizing 108 Prior A nn u. R ev . P ol it. S ci . 2 01 3. 16 :1 01 -1 27 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g by P ri nc et on U ni ve rs ity L ib ra ry o n 05 /1 4/ 13 . F or p er so na l u se o nl y. PL16CH06-Prior ARI 7 April 2013 1:28 audience attitudes, slanted news may thus move everyone in the same direction, especially when audience members lack political knowledge or strong prior attitudes. As Chong & Druckman (e.g., 2007) demonstrate, the strength of arguments also affects the likelihood of persuasion. Partisan media slant is not a necessary condition for attitude polarization either. In both the RAS model and the motivated reasoning framework, exposure even to balanced or neutral news can lead to attitude polarization (see also Lord et al. 1979). In the RAS model, this happens when recipients can use source cues to differentiate conservative and liberal arguments within a balanced newscast or when they reject counter-attitudinal arguments based on prior knowledge. Zaller’s (1992, 1996) work provides many examples of attitude polarization during times when slanted news was rare. In themotivated reasoning framework, balanced news can polarize attitudes when ideological recipients dismiss and counter-argue attitude-inconsistent information while embracing attitude-consistent information. This makes it empirically difficult to distinguish the impact of more partisan media from the impact of more polarized parties using balanced media to convey information about their platforms. In short, attitude polarization is one theoretically plausible, empirically demonstrated outcome of media exposure—but not the only one. The impact of partisan media on those who follow them depends in complex ways on their preexisting attitudes and political sophistication. Moreover, the mechanisms of media influence discussed so far—attitude polarization and persuasion—presume exposure. Yet, the proliferation of news outlets and platforms gives people greater control over theirmedia exposure and thus greater opportunity to avoid counter-attitudinalmessages (Festinger 1957, Sears & Freedman 1967, Frey 1986). This points to a weakness in Zaller’s (1992, p. 139) model: he rules out partisan selective exposure, stating, “It is the implicit assumption that a person’s predispositions, although affecting acceptance of persuasive messages, do not affect reception.” Motivated reasoning, in contrast, covers the possibility of partisan selectivity (e.g., Taber &Lodge 2006). In studies of motivated reasoning, messages are often attributed to politically unequivocal sources (such as parties or candidates affiliated with parties). A growing number of studies aim to gauge the extent of partisan selective exposure in the current high-choice media environment where many centrist sources remain. Partisan Selective Exposure Several experimental studies have manipulated media source cues to document partisan selective exposure. Iyengar&Hahn (2009) presented subjectswith four news headlines. Randomly assigned, some subjects also saw the logo of a news organization next to each headline (Fox News, NPR, CNN, BBC). Among Republican participants, adding the Fox News logo to a headline increased by about 25 percentage points the chance that participants would want to read the story. Adding the CNN or NPR logo reduced the probability by close to 10 points. Among Democrats, the effects were smaller, with a reduction in selection of just over 10 points when a headline was labeled “Fox News.” [Messing & Westwood (2011) obtain very similar results.] Unobtrusive observation can document selection behavior even in the absence of a control group. Stroud (2011, pp. 67–73) offered people several news magazines while they were waiting for the supposed beginning of a research study and observed their choices. Participants were marginally more likely to read ideologically consistent magazines and select a free subscription to ideologically consistent magazines at the end of the study. Iyengar et al. (2008) automatically tracked use of amultimediaCDwithmaterial generated by the two presidential campaigns in 2000. Study participants were significantly more likely to look at content about personally relevant issues. Strong Republicans and conservatives were more likely to access Bush content, but the study generates no evidence for selective exposure among strong Democrats. Iyengar et al. (2008, www.annualreviews.org • Media and Political Polarization 109 A nn u. R ev . P ol it. S ci . 2 01 3. 16 :1 01 -1 27 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g by P ri nc et on U ni ve rs ity L ib ra ry o n 05 /1 4/ 13 . F or p er so na l u se o nl y. PL16CH06-Prior ARI 7 April 2013 1:28 p. 197) call the overall support for partisan selective exposure “modest” and conclude that “our participants’ exposure decisions were driven more by generic interest in politics and interest in particular issues, so there was plenty of exposure to potentially distasteful information.” Using a similar design, Kim (2009) also finds evidence that access to detailed candidate information promotes issue-based selective exposure (which reduces the impact of partisan considerations in candidate evaluations). In Graf & Aday’s (2008) study, student subjects spent more time reading online news about their preferred candidate or party but accessed counter-attitudinal stories on their preferred issues aswell. In a disproportionately partisan sample (Garrett 2009a), subjects were more likely to express interest in reading news stories that they considered opinion-confirming based on headline, source, and a short preview. However, their tendency to avoid stories expected to be opinion-challenging was only marginally significant, and they spent as much time reading stories they expected to be opinion-challenging. Bennett& Iyengar (2008, p. 722) interpret the difference between the studies conducted in 2000 (Iyengar et al. 2008) and 2006 (Iyengar & Hahn 2009) as a genuine trend toward more selective exposure. But an experiment conducted by Stroud (2011, pp. 73–77) in 2008 that resembles the Iyengar & Hahn design also found stronger evidence for issue-based selectivity than for partisan selectivity. Moreover, Bennett & Iyengar (2008) compare apples (selective exposure to different candidate-generated content attributed to political sources) and oranges (selective exposure to otherwise identical newsheadlines attributed todifferentmedia sources). It is not clear theoretically or empirically if selection operates in the same way in both contexts. Messing &Westwood (2011) extend the Iyengar&Hahn design by adding social endorsement cues (“10,000 people recommend this story”) next to the story headlines. These cues override the impact of source labels, leading people to select news from ideologically incongruent media outlets. There is some evidence, then, that some people, when given a choice between news reports with different (expected) partisan slant, gravitate toward like-minded news, but other selection criteria also operate and often override partisan consistency. It is possible that partisan selectivity varies at the individual level and depends on the task. In Valentino et al. (2009), subjects who were made to feel anxious about the 2004 presidential campaign sought counter-attitudinal candidate information more than those made to feel enthusiastic (unless subjects were told they would later have to defend their candidate preferences). Outside the experimental context, people face a choice of content over and over again. When choosing repeatedly, selective exposure might decline (Fischer et al. 2005), and even partisans may sometimes select a centrist or counter-attitudinal news source. In an experiment in which subjects chose twice between news stories that were distinguished only by randomly assigned source information, a substantial share of participants used their second choice to ideologically balance their first one.2 In a more general criticism of experimental work, Iyengar et al. (2008, p. 187) note that “virtually all controlled research to date on selective exposure has relied on nonspontaneous or limited information search situations.” With respect to Internet use, the combination of automatic tracking with survey data on user ideology makes it possible to observe natural, “spontaneous” media use—although at the expense of control over the choice set. Analyzing comScore Internet tracking data for a sample of participants who are also interviewed to measure their ideology, Gentzkow & Shapiro (2011) find weak evidence for partisan selectivity in online media use. Most of the largest news websites attract a similar amount of traffic from conservative, moderate, and liberal users. 2This experiment was conducted undermy supervision byChristopher Sykes for his senior thesis at Princeton in 2010. Iyengar & Hahn (2009) asked each subject to make six selection decisions. Their subjects did not actually see the selected content, however, and Iyengar & Hahn only report average treatment effects. 110 Prior A nn u. R ev . P ol it. S ci . 2 01 3. 16 :1 01 -1 27 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g by P ri nc et on U ni ve rs ity L ib ra ry o n 05 /1 4/ 13 . F or p er so na l u se o nl y. PL16CH06-Prior ARI 7 April 2013 1:28 Thirty percent of daily visitors to http://www.nytimes.com are conservatives, for example, as are 22% of visitors to http://www.huffingtonpost.com. Some websites do attract mostly conservatives or mostly liberals. Three-quarters of daily http://www.foxnews.com visitors are conservatives. Audiences for smaller opinion websites, such as http://www.billoreilly.com or http://www.rushlimbaugh.com, are even more skewed ideologically. Yet even if audiences for some small websites in the long tail are ideologically pure, most of their users are not, and even ideologues tend to visit an ideologically diverse selection of websites. “Their omnivorousness outweighs their ideological extremity, preventing their overall news diet from becoming too skewed” (Gentzkow & Shapiro 2011, p. 1,832). For example, 30% of monthly visitors to http://www.rushlimbaugh.com also visit http://www.nytimes.com in the same month, and 50% also visit Yahoo! News. A quarter of monthly http://www.dailykos.com visitors also visit http://www.foxnews.com. Overall, Gentzkow & Shapiro (2011) estimate ideological segregation, which is conceptually similar to “de facto selectivity” (Sears & Freedman 1967, p. 197; see also Mutz & Young 2011), to be only slightly larger online than in the average county and slightly smaller than in the average ZIP code. Studies of selective exposure on television typically reach a very different conclusion: Republicans and conservatives report more exposure to conservative outlets, whereas Democrats and liberals report greater exposure to liberal sources, so selective exposure in cable news viewing is common (e.g., Coe et al. 2008, Hollander 2008, Jamieson & Cappella 2008, Garrett 2009b, Stroud 2011, Holbert et al. 2012). Hollander (2008, p. 34), for example, argues that “Republicans over time have shifted to such sources as Fox News, generally seen as friendly to their beliefs, and away from news sources often named by conservative critics as unfriendly to their side, such as CNN. . . . Democrats have done the same.” Stroud (2011, p. 54) concludes from her analysis of survey data that “political beliefs seem to very much influence cable news network preferences.” Jamieson & Cappella (2008, p. 81) write that “the audiences of the conservative media establishment are disposed to hold attitudes, opinions, and ideology that agree with these media sources. The conservative media establishment is ‘preaching to the choir.’” There are two plausible explanations for the divergence between these studies and Gentzkow & Shapiro’s (2011) analysis of Internet use. Either cable viewers are in fact a lot more selective in deciding which partisan slant to follow than Internet users, or a methodological difference accounts for the results. Unlike Gentzkow & Shapiro, who examine Internet tracking data, the cable news studies cited above all rely on media use self-reported by survey respondents. The next section describes the cable news audience using automatic tracking data to understand if selectivity in cable news viewing is as significant as many survey-based studies indicate. HOW MANY PEOPLE WATCH PARTISAN NEWS? To properly assess the influence of partisan television, it is critical to know howmany people watch partisan media, for how long they watch, who they are, and what else they watch (or read or hear.) The main source of tracking data for US television use is The Nielsen Company, which monitors television viewing in a random sample of about 10,000 US households to derive proprietary audience estimates (for more information about Nielsen’s methodology, see Prior 2009a; Napoli 2003, pp. 71–95). Detailed Nielsen estimates (let alone individual-level data) are not publicly available, however, so characterizations of the cable news audience are patchy. A second source of tracking data ( Jackman et al. 2012, LaCour 2012) relies on a different technology (cell phones picking up radio and television sound, which can be matched against a database of programming) and does offer individual-level data, but only covers the New York and Chicago media markets. Jointly, these two data sources do help us understand the cable news audience better. www.annualreviews.org • Media and Political Polarization 111 A nn u. R ev . P ol it. S ci . 2 01 3. 16 :1 01 -1 27 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g by P ri nc et on U ni ve rs ity L ib ra ry o n 05 /1 4/ 13 . F or p er so na l u se o nl y. PL16CH06-Prior ARI 7 April 2013 1:28 Nielsen’s most easily available estimates of cable news viewing are ratings. Over most of the past decade, the Fox News Channel (FNC) has received the highest ratings of all cable news channels. Its average primetime audience in recent years has been around two million, or 0.7% of the population.The highest-rated hour onFNC (and all of cable news),TheO’Reilly Report, attracts on average about three million viewers. The average primetime audience for FNC, CNN, and MSNBC combined is about 1.1% of the population (or 1.4% of adults) (Project for Excellence in Journalism 2012). Yet ratings give a somewhat misleading impression of the cable audience because they average over long portions of the day when most people are asleep or at work. Even the average audience for individual shows in primetime obscures the number of regular viewers because it only gives full weight to someone who watches the entire show every day it airs (see Prior 2007, pp. 151–55 for elaboration). A more relevant metric is the cumulative audience (also referred to as “reach” or “cume”). It is an estimate of the number of unique people who watch a particular program or channel for more than a certain number of minutes in a specified period. For example, 163 million people in the United States watched some or all of the Super Bowl in 2011, but the average minute was seen by only 111 million people (FOX 2011). Even though FNC has consistently attracted a greater average audience than CNN during the past decade, CNN has equally consistently had more unique viewers. In other words, the number of people who ever watch CNN exceeds the number who watch FNC, but FNC viewers tune in for longer periods of time than CNN viewers. The average monthly one-minute cumes have been around 100 million for CNN and 80 million for FNC in recent years (Project for Excellence in Journalism 2011). These numbers, too, are misleading because they include many cable news viewerswho tune in extremely briefly. A comparison of one-minute and six-minute cumes indicates that 35% of CNN viewers and 30% of FNC viewers watch less than five minutes per month. (Nielsen includes exposure to commercials, so these are essentially nonviewers.) According to six-minute cumes, about a quarter of the population watched CNN for at least six minutes per month between 2004 and 2009. For FNC, this share was about a fifth. With a threshold of six minutes per month, these estimates include even the most sporadic viewers. For 2010, the Project for Excellence in Journalism also published estimates with a higher threshold, the monthly 60minute cume. An average of about 41.7 million viewers watched CNN for at least one hour per month, and another 14.4 million watched between six and 50 minutes. For FNC, the equivalent audiences were 41.0 million and 10.0 million, respectively. But even 60 minutes of exposure to a mix of cable news and commercials over the course of a month does not amount to a heavy dose of news. Monthly cumes with higher thresholds have not been published, but Stroud (2011, pp. 208–10) presents estimates for several two-week periods in 2004 and 2008, and I obtained cume data for one week in March 2008. These data represent the best available information on the concentration of news viewing and the share of heavier news viewers. In April 2008, 22.7 million people watched FNC for at least an hour over a two-week period, and 17.8 million watched for at least two hours. And during the week of March 22, 13.5 million watched FNC for an hour or more. The first estimate includes people who watch an average of just over four minutes of FNC programming and commercials per day, and even so, only 10% of the adult population is in this group.3 The second and third estimates amount to an average 3Stroud’s estimates are for the entire population (P2+).When expressing them as a percentage of the adult population (P18+), I thus assume that cable news viewing in the 2–17 age group is minimal, an assumption consistent with the typical age gradient of TV news consumption (see Prior 2009a, 2012). To the extent that the assumption is incorrect, the adult news audience is even smaller. Data for the week of March 22 are for P18+. 112 Prior A nn u. R ev . P ol it. S ci . 2 01 3. 16 :1 01 -1 27 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g by P ri nc et on U ni ve rs ity L ib ra ry o n 05 /1 4/ 13 . F or p er so na l u se o nl y. PL16CH06-Prior ARI 7 April 2013 1:28 daily minimum of 8.5 minutes per day, and only 6.1–7.9% of adults watch this much FNC. (That the two-hour cume over two weeks is only slightly larger than the one-hour cume over one week suggests that most of these heavier viewers regularly watch an hour or more per week.) These magnitudes appear to be fairly typical. Stroud’s data for April and October 2004 show nearly identical oneand two-hour cumes for FNC (one-hour: 22.5 million and 24.8 million viewers; two-hour: 16.5 million and 18.8 million). CNN and MSNBC have smaller audiences at this level of exposure. Sixty-minute cumes were 4.8% of adults for CNN and 3.8% forMSNBC in the week of March 22, 2008. In Stroud’s fourth observation period, October 2008—the final month of the Obama-versusMcCain campaign in themidst of the economic collapse—FNCexposurewas higher.Themonthly six-minute cume for October 2008 was 63 million, 16% higher than in March/April; the biweekly 60-minute cumewas 28%higher and the biweekly 120-minute cume 26%higher. And the average primetime audience for FNC nearly doubled between March/April and October 2008. Hence, Fox News consumption increased sharply as Election Day 2008 approached partly because more people tuned in but more because existing FNC viewers spent more time watching. The pattern is even stronger for CNN and MSNBC. Primetime viewing more than doubled, but the share of the population that watched at least briefly increased at a lower rate.4 By any reasonable definition of heavy news viewing, then, audiences for FNC, CNN, and MSNBC are small. The share of Americans who watch cable news at a rate of 10 minutes or more per day is probably no larger than 10–15% of the voting-age population and rises modestly when an exciting election approaches. Even this estimate may be high because adding up separate cume estimates for each cable news channel amounts to double-counting people who watch more than one channel. Data on the overlap between audiences for different news channels provide additional information about the share of heavy viewers and the extent of selective exposure in cable news viewing. Do Fox News Viewers Tune Out the Other Side? Because Nielsen does not measure the party ID or ideology of panelists, its data cannot reveal the partisan composition of the audience. Yet audience overlap, the share of viewers of one channel who also watch another, provides some insight into selective exposure. Minimal overlap between audiences for conservative and liberal channels would correspond to strong selective exposure. Tracking data on the overlap between audiences for different channels are rarely published. Prior (2007, pp. 156–58) uses Nielsen data obtained by Webster (2005) for February 2003 and reports strong signs of overlap: the average FNC viewer, defined as anyone who watched at least one minute per week, devoted 7.5% of her overall news consumption to FNC but another 6% to other cable news channels. CNN viewers spent 4.7% of their viewing time watching CNN but also 3.7% watching FNC. By using a one-minute weekly cume to define viewers of a channel at a time when the imminent military intervention in Iraq was driving up news audience numbers overall, these data effectively characterize average audience overlap in a sample that includes a few heavy viewers and many more individuals who watched for only a few minutes. Among heavier viewers, audience overlap appears to be less common. In Stroud’s (2011) data, 65% of people who watched FNC for an hour or more over two weeks also watched a mix of 4Between March/April and October 2008, primetime viewing increased by 119% for CNN and 124% for MSNBC. The six-minute monthly cume over the same period rose by 18% for CNN and 3% for MSNBC. The smaller rise in primetime viewing in 2004 was also concentrated among existing viewers: primetime viewing increased by 25% for CNN, 53% for MSNBC, and 60% for FNC. The six-minute monthly cume over the same period increased 15% for CNN, 8% for MSNBC, and 19% for FNC. www.annualreviews.org • Media and Political Polarization 113 A nn u. R ev . P ol it. S ci . 2 01 3. 16 :1 01 -1 27 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g by P ri nc et on U ni ve rs ity L ib ra ry o n 05 /1 4/ 13 . F or p er so na l u se o nl y. PL16CH06-Prior ARI 7 April 2013 1:28 CNN and MSNBC for at least 15 minutes during the same two weeks. But of FNC viewers who watched at least two hours, just 45% also watched the other cable news channels. These ratios are remarkably stable across the four periods for which Stroud presents data. They demonstrate that heavier FNC viewers are less likely to watch CNN or MSNBC at least briefly. (The relationship between duration of exposure to CNN orMSNBC and avoiding FNCmay be somewhat weaker.) Overlap is less frequent yet when it is defined as watching two channels for at least an hour per week each. Of those who watched at least an hour of FNC in the week of March 22, 2008, only 16% also watched at least an hour of CNN and 13% at least an hour of MSNBC. Only 20% of those who watched CNN for at least an hour, and 21% of those who watchedMSNBC for at least an hour, also watched FNC for an hour or more. When the six-minute cume is used instead to examine a larger but less intense news audience, overlap increases. Of those who watched at least six minutes of FNC, 42% also watched at least six minutes of CNN and 33% at least six minutes of MSNBC. There are thus clear signs of selective exposure among some heavy cable news viewers. It is important to note, however, that this is a small group. Only 5% of the adult population watched an hour or more of FNC in the week of March 22, 2008 but less than an hour of CNN. Because this figure does not considerMSNBC exposure, it likely overstates selective exposure. LaCour’s (2012) analysis of news exposure in the New York and Chicago media markets in the 12 weeks before the 2006 midterm elections includes exposure to other partisan television and radio programs. Only 7% of his sample devoted 80% or more of their news exposure to one side. All but the most selective cable news viewers tend to watch a lot of (ideologically heterogeneous, often neutral) local news. The data used by LaCour (2012) have another advantage over Nielsen because they include the partisan affiliation of their panelists. LaCour can therefore examine selective exposure more directly by comparing news exposure of Democrats and Republicans. On average, Democrats selected 23% liberal news and 10% conservative news. For Republicans, 22% of media use was devoted to conservative news and 15% to liberal news. The large majority of partisans followed mostly local news. Like-minded exposure exceeded 80% of total news exposure among only 4% of partisans. Automatic tracking of television viewing using two different technologies reveals that most people avoid cable news almost entirely. A large segment watches cable news infrequently and nonselectively, mixing exposure to different cable news channels. In the small slice of heavy cable news viewers, however, partisan selective exposure is not uncommon. Measurement Error and Partisan Selective Exposure Because asking people about their news exposure in surveys is a lot cheaper than tracking their viewing automatically, a question of critical importance for designing future research is whether survey-based self-reports can accurately categorize respondents as nonviewers, nonselective casual viewers, nonselective heavy viewers, or selective heavy viewers. Existing research indicates that many self-reports ofmedia use are not very accurate. Low validity has been shown for self-reported exposure to network news (Prior 2009a), all television news (LaCour 2012), and presidential debates (Prior 2012). The main cause of inaccurate self-reports appears to be the failure to recall exposure in any detail. Asked to report it anyway, survey respondents give inaccurate estimates that often inflate their actual exposure (Prior 2009b). If anything, theories of motivated reasoning heighten validity concerns for exposure to partisan media: taking self-reports at face value requires the assumption that the very peoplewho follow theirwishful thinkingwhen they evaluate economic performance (see Bartels 2002) or perceive centrist news as biased (see Dalton et al. 1998, Vallone 114 Prior A nn u. R ev . P ol it. S ci . 2 01 3. 16 :1 01 -1 27 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g by P ri nc et on U ni ve rs ity L ib ra ry o n 05 /1 4/ 13 . F or p er so na l u se o nl y. PL16CH06-Prior ARI 7 April 2013 1:28 et al. 1985) faithfully report when they turn off the pro-attitudinal message stream or follow counter-attitudinal programming. If instead partisans forget, underestimate, or fail to admit their exposure to “the other side,” but are happy to report following “their side,” self-reports will exaggerate the extent of selective exposure. To compare automatic tracking results to survey-based estimates, I draw on two surveys conducted in the first half of 2008. TheMedia Consumption Survey conducted biannually by the Pew Research Center for the People and the Press is arguably the most detailed and thorough survey on news media use. The 2008 Media Consumption Survey (n = 3,615) was conducted by phone between April 30 and June 1. The survey used random-digit dialing and included cell phone users in its sample. The second survey was conducted by KnowledgeNetworks (KN) betweenMarch 26 and April 8, 2008 (n = 1,583). Respondents constitute a randomly selected subset of KN’s larger panel, which is recruited through probability sampling. (Like Nielsen estimates, both surveys use demographic post-stratification weights.) Self-reports exaggerate the size of the cable news audience. In the Pew survey, 52% of Americans aged eighteen and older reported watching Fox News “regularly” or “sometimes.” In the KN survey, 55% did. The share of self-proclaimed “regular” Fox News viewers was 24% according to Pew and 22% in the KN survey.5 The vague response options in these and most other survey questions about cable news exposure prevent an exact comparison to Nielsen estimates. But no matter which cume threshold is used, Nielsen’s estimates are much smaller, indicating considerable overreporting in surveys. Only 6–8% of Americans watched at least 60 minutes of FNC per week in March/April 2008, but three to four times as many called themselves regular viewers of the news channel. Even if we define a regular viewer as anyone who watches more than five minutes per week, which is surely too lenient a definition, survey estimates are still twice as high as Nielsen estimates.Onewould have to implausibly treat sixminutes of exposure per month as regular viewing in order to achieve rough correspondence between self-reports and Nielsen estimates. Overreporting of exposure to CNN and MSNBC is similar in magnitude. Fewer than 4% of American adults watched an hour of MSNBC or more in a week, but three times as many called themselves regular viewers. The 60-minute cume for CNNwas 4.8%, compared to 18% “regular” CNN viewers, averaged across the two surveys. If audience estimates based on automatic tracking are so much smaller than estimates that rely on self-reports, conclusions drawn from survey data about audience overlap should also raise suspicion. Figure 1 presents estimates of audience overlap between FNC andMSNBC inNielsen data, the Pew survey, and the KN survey. The figure illustrates the general mismatch between self-reports andNielsen estimates. Self-reports, especially those that include “sometimes” viewers, produce dramatically higher estimates of audience size and amount of overlap. Of survey respondents who report watching FNC at least “sometimes,” 57% (KN) and 58% (Pew) also report “sometimes” or “regularly” watching MSNBC. By comparison, only 33% of people who watch at least six minutes of FNC per week also watch six minutes or more of MSNBC. The mismatch is slightly greater for overlap between FNC and CNN. Audience overlap is lower among self-reported regular viewers. Only 22% (KN) or 25% (Pew) of “regular” FNC viewers also watch MSNBC “regularly.” Both survey and tracking estimates 5Pew uses the following question wording: “Now I’d like to know how often you watch or listen to certain TV and radio programs. For each that I read, tell me if you watch or listen to it regularly, sometimes, hardly ever, or never. How often do you watch [program or channel]?” The exposure questions in the KN survey were designed to match the Pew format. Respondents were randomly assigned to a series of questions about either cable news channels or cable news programs. The top of the survey screens read, “How often do you watch each of these cable news channels [programs]?” The response options were the same as in the Pew survey. www.annualreviews.org • Media and Political Polarization 115 A nn u. R ev . P ol it. S ci . 2 01 3. 16 :1 01 -1 27 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g by P ri nc et on U ni ve rs ity L ib ra ry o n 05 /1 4/ 13 . F or p er so na l u se o nl y. PL16CH06-Prior ARI 7 April 2013 1:28