Public Opinion Quarterly, Vol. 76, No. 3, Fall 2012, pp. 405-431 AFFECT, NOT IDEOLOGY A SOCIAL IDENTITY PERSPECTIVE ON POLARIZATION SHANTO IYENGAR* GAURAV SOOD YPHTACH LELKES Abstract The current debate over the extent of polarization in the American mass public focuses on the extent to which partisans' policy preferences have moved. Whereas "maximalists" claim that partisans' views on policies have become more extreme over time (Abramowitz 2010), "minimalists" (Fiorina and Abrams 2009) contend that the majority of Americans remain centrist, and that what little centrifugal movement has occurred reflects sorting, i.e., the increased association between partisanship and ideology. We argue in favor of an alternative definition of polarization, based on the classic concept of social distance (Bogardus 1947). Using data from a variety of sources, we demonstrate that both Republicans and Democrats increasingly dislike, even loathe, their opponents. We also find that partisan affect is inconsistently (and perhaps arti-factually) founded in policy attitudes. The more plausible account lies in the nature of political campaigns; exposure to messages attacking the out-group reinforces partisans' biased views of their opponents. Introduction Scholarship on political polarization, with very few exceptions, focuses exclusively on policy preferences. By this definition, American party elites have become increasingly polarized over the past four decades (Fleisher and Bond 2001; Hetherington 2002; McCarty, Poole, and Rosenthal 2006). Roll Shanto Iyengar is the Harry & Norman Chandler Professor of Communication and Professor of Political Science, Stanford University, Stanford, CA, USA. Gaurav Sood is apostdoctoral research associate, Woodrow Wilson School of Public and International Affairs, Princeton University, Princeton, NJ, USA. Yphtach Lelkes is an Assistant Professor of Political Communication, Amsterdam School of Communication Research, The University of Amsterdam, The Netherlands. Shanto Iyengar's contribution to this work was supported by the National Research Foundation of Korea Grant funded by the Korean government [NRF-2010-330-B00028]. *Address correspondence to Shanto Iyengar, Stanford University, Building 120, Room 110, 450 Serra Mall, Stanford, CA 94305, USA; e-mail: siyengar@stanford.edu. doi:10.1093/poq/nfs038 Advance Access publication 17 September 2012 © The Author 2012. Published by Oxford University Press on behalf of the American Association for Public Opinion Research. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 406 Iyengar, Sood, and Lelkes call analysis of congressional voting (Poole, Rosenthal, and Koford 1991; McCarty, Poole, and Rosenthal 2001), interest-group ratings of congressmen (Stonecash, Brewer, and Mariani 2003), analysis of party platforms (Layman 1999), and surveys of party activists (Aldrich 1995; Layman 1999; Layman, Carsey, and Horowitz 2006) all show increased division between the Democrats and the Republicans on the issues. Some argue that elite polarization is not a response to mass polarization (Fiorina, Abrams, and Pope 2005), but is instead an (unintended) consequence of institutional changes (Burden 2001).1 Relatedly, they argue that the elites haven't persuaded their followers to take on more extreme positions (Levendusky 2009). These scholars portray the median citizen as someone who is centrist on most issues. Others contest this description of the masses, citing a decline in the number of ideological moderates (Abramowitz and Saunders 2008; Campbell 2008; Jacobson 2000), and a near doubling of the average distance between the ideological self-placement of non-activist Democrats and Republicans between 1972 and 2004 (Abramowitz and Saunders 2008). In short, there is considerable debate over whether the masses are ideologically polarized (for reviews of the evidence, see Fiorina and Abrams [2009] and Hetherington [2009]). Policy-based division is but one way of defining partisan polarization. An alternative, and in our view, more diagnostic, indicator of mass polarization is the extent to which partisans view each other as a disliked out-group. The definitional test of social identity (Tajfel 1970; Tajfel and Turner 1979) requires not only positive sentiment for one's own group, but also negative sentiment toward those identifying with opposing groups. So, to the extent that party identification represents a meaningful group affiliation, the more appropriate test of polarization is affective, not ideological, identity.2 Affect based on partisan identity may reflect some combination of ideological disagreements and group-centric impulses. Partisans may attribute the more extreme policy positions of party elites and activists to party supporters, producing "principled" dislike. We present evidence that partisan affect is inconsistently related to policy preferences and that the relationship between partisan affect and policy attitudes hasn't notably strengthened over time. The 1. Other proposed explanations of elite polarization that exculpate the mass public include the changing basis of party activism, from clientelist to ideological (Fiorina, Abrams, and Pope 2005), the rise of ideological interest groups (Scholzman and Tierney 1986), changing composition of electoral districts (Stonecash, Brewer, and Mariani 2003), and the decline of Southern Democrats in the post-civil rights era (Rohde 1991). 2. Some of the scholars who have incorporated affect into the analysis of polarization include Hetherington and Weiler (2009), who demonstrated that authoritarianism contributed to higher ratings of the Republican Party by Republicans and indirectly to lower ratings of the Republican Party by Democrats (pp. 150-53), and Richardson (1991), who argued that European party systems that reflect long-standing social, cultural, and religious cleavages generate more intense conflict across party lines (Richardson 1991). A Social Identity Perspective on Polarization 407 more plausible explanation, in our view, is that the mere act of identifying with a political party is sufficient to trigger negative evaluations of the opposition, and exposure to prolonged media-based campaigns only reinforces these predispositions. We present evidence showing that partisans' feelings for their opponents become more negative following exposure to presidential campaigns. Our principal goal is to document the extent of affective polarization. We show that Democrats and Republicans not only increasingly dislike the opposing party, but also impute negative traits to the rank-and-file of the out-party. We further demonstrate that affective polarization has permeated judgments about interpersonal relations, exceeds polarization based on other prominent social cleavages, and that levels of partisan affect are significantly higher in America, compared to the United Kingdom. Finally, we present evidence suggesting that exposure to political campaigns is a potential contributory factor. Overall, the evidence is unequivocal: in terms of affect, Americans are polarized along party lines. The article is organized as follows. We begin with a discussion of social identity theory and the prediction of enhanced social distance between partisans. We then describe the data used to measure affective polarization. This is followed by a results section showing (1) increased polarization over time within the United States, (2) increased divergence between the United States and the United Kingdom (a suitable "control" nation) in the extent of affective polarization, (3) that voters' policy preferences are only weakly associated with partisan affect, and (4) that exposure to political campaigns in general and negative advertising in particular strengthens partisan affect. In closing, we discuss other potential explanations for heightened affective polarization, and discuss some of the implications for democratic politics. Theoretical Framework Group dynamics have traditionally been at the forefront of social science research. Social psychology experiments demonstrate that any form of group membership, even one based on the most trivial of shared characteristics (e.g., the tendency to overestimate the number of dots), triggers both positive feelings for the in-group, and negative evaluations of the out-group (Tajfel et al. 1971; Billig and Tajfel 1973). These findings suggest that group-based affect is an ingrained human response, occurring even when group membership is assigned randomly (Billig and Tajfel 1973). Since individuals typically categorize themselves into multiple groups, an important question concerns the hierarchy of group affiliations. In contexts involving multiple groups, which affiliations provide the most meaningful cues? Social identity theorists posited identity salience as the basis from which to predict the extent of inter-group prejudice (see Oakes 1987 and Gaertner 408 Iyengar, Sood, and Lelkes et al. 1993): the more salient the affiliation, the more biased the individual's beliefs about in-group and out-group members. Salience itself can depend on either dispositional factors, such as the strength of the individual's loyalty to the group, or characteristics of the information environment, such as the number of times the individual is reminded of her affiliation to some group. In the case of party identification, for instance, we might expect strong partisans to carry more biased perceptions of their opponents. As we argue below, it can also be expected that exposure to political campaigns serves to heighten the salience of partisan identity among all identifiers. REVISITING FEAR AND LOATHING ON THE CAMPAIGN TRAIL Despite its title, Hunter S. Thompson's (1973) classic account of the 1972 presidential campaign described, at least by contemporary standards, a relatively soft-spoken campaign. By any relevant measure—length of time, amount of campaign finance, scope and reach of television ads, volume of news media coverage—American presidential campaigns have since become more antagonistic and harder to ignore. There can be no doubt about the increased negativity of campaign rhetoric. Candidates routinely spend more time attacking their opponents than promoting themselves (Geer 2010). Negative (but not positive) messages are recycled ad infinitum by journalists who seek conflict and controversy above all else. Coverage of advertising has become a staple feature of news, so much so that the 2004 Swift Boat ad impugning Senator Kerry's Vietnam record generated more news stories than the war in Iraq (see Geer 2010). The availability of news coverage of the campaign has risen steadily since the advent of cable television networks. On Fox News and, more recently, MSNBC, partisans encounter news and commentary on the campaign that only confirms their stereotypes of the out-group. Access to campaigns was enhanced further in the post-Internet era with news reports playing on laptop computers and cell phones on a daily basis. Thus, greater exposure to the harsh rhetoric of political campaigns is one potential explanation of affective polarization. Research Design We rely on data from national and cross-national surveys to assess whether partisans' dislike of each other has risen over the past fifty years in the United States, and how the current and past levels of inter-partisan conflict compare with those in the United Kingdom. First, we describe the trend in the feeling thermometer ratings of the in- and out-parties in the United States. Second, using multiple surveys conducted between 1960 and 2010, we describe parallel changes in two indicators of inter-party social distance—stereotypes of party supporters and feelings about marriage across party lines. A Social Identity Perspective on Polarization 409 Next, we present longitudinal and contemporaneous comparisons between the United States and the United Kingdom. The UK provides us with a relatively stringent baseline because British parties have strong roots in the class system. Since the sense of party identity reflects both ideological and social cleavages, we should anticipate higher levels of inter-partisan affect in the United Kingdom. Comparing the two nations over time allows us to assess whether the trend in the United States reflects a general over time trend or is a country-specific effect. We then compare party affiliation with other prominent social cleavages in the United States, including gender, race, and religion, as bases for affective polarization. These data show that negative affect for the out-group is the greatest when group identity is defined in terms of partisanship. Having documented significant increases in affective polarization, we proceed to consider two possible explanations for this trend. First, we explore whether affective polarization is but a consequence of ideological polarization. The results show only a modest association between economic policy preferences and views of the out-party, but little change over time in the strength of this relationship. Second, we use two different national surveys and data on campaign advertising expenditures to show that exposure to negative advertising in 2004 and to the overall campaign in 2008 contributed to increased partisan animus. DATA SETS We take advantage of six different survey data sets—American National Election Studies (ANES), a 2008 YouGov poll of U.S. and UK voters, the Almond and Verba 1960 five-nation study that formed the basis of The Civic Culture, a 2004 Blair Center Election Study, the U.S. component of an eleven-nation study conducted by YouGov in 2010, and the AP-Yahoo! News 2008 election study. ANES: We take advantage of the time series, and the 1988 and 2004 cross-sectional surveys. Since 1948, ANES has conducted national surveys measuring political attitudes and behaviors using mostly in-person interviews. The time series consists of both pre- and post-election interviews. Although declining over time, the AAPOR 3 response rate for the ANES surveys has ranged between 60 and 71 percent. YouGov/Polimetrix: (1) In March 2008, the Economist and the Hoover Institution commissioned a survey of voters in the United States and the United Kingdom. YouGov samples from a pool of opt-in respondents using a two-stage procedure. A conventional probability sample is drawn using a large-scale RDD sample (the target sample). Next, for each member of the target sample, YouGov selects one or more matched members—matching on race, gender, age, education, and imputed party identification—from their pool of opt-in respondents. The end result is a sample of opt-in respondents with equivalent characteristics as the target sample on the matched characteristics (Rivers and Bailey 2009). The pooled—United States and UK combined— AAPOR 1 response rate was 42.1 percent. (2) As part of an eleven-nation 410 Iyengar, Sood, and Lelkes comparative study of citizen knowledge, YouGov surveyed a sample of adult Americans between March and April 2010. The sampling procedures are as described above except for the fact that there was an oversample of households without cable television. The AAPOR 1 response rate was 42 percent. Almond and Verba: Between 1959 and 1960, multistage probability samples of adults from the United States and the United Kingdom were interviewed as part of the Five-Nation Civic Culture Study (Almond and Verba 1960). The AAPOR 1 response rate in the United States was 83.3 percent, while it was 59 percent in the UK. 2004 Blair Center Election Study: This Knowledge Networks survey was fielded immediately after the 2004 presidential campaign (see Hillygus and Shields 2004); 2,837 respondents (1,168 from Southern states, 1,166 from other states, 503 from general adult population) were interviewed between November 5 and 16, 2004. The overall survey completion rate was 68.1 percent. 2008 Associated Press-Yahoo! News Election Panel Study: This survey, administered by Knowledge Networks, included eleven waves of data and was fielded between November 2007 and December 2008.3 The cumulative response rate across the first nine waves (used in the analyses) was 9.8 percent. Wisconsin Advertising Project: The data set contains ads broadcast on the national broadcast and cable television networks in the top 100 media markets during the 2004 elections, as tallied by the Campaign Media Analysis Group.4 The investigators coded advertisements for a variety of characteristics including negativity (see Goldstein and Rivlin 2007a, 2007b). Measures THERMOMETER RATING OF PARTIES, PARTISANS, AND IDEOLOGICAL GROUPS ANES: Respondents rated different groups on a thermometer scale5 ranging from 0 to 100, where a score of 0 means the respondent feels "cold" toward the group, a score of 50 means the respondent either doesn't know much about the group (before 1968) or doesn't feel "particularly warm or cold toward" the group (after 1968), and a score of 100 implies the respondent has "warm" feelings toward the group. Intervening numbers can be seen as interpolations between these semantic points. 3. More information about the study methodology can be obtained at http://www.knowledgenet-works.com/ganp/election2008/index.html. 4. We rely on 2004 data instead of 2008 because of the unavailability of presidential ad data for 2008. 5. The precise wording of the question and some of the directions on how to use the scale have changed overtime. See ANES Guide 2010, available at http://www.electionstudies.org, for details. A Social Identity Perspective on Polarization 411 The ANES has asked the partisan thermometer in two different ways over the years. Between 1964 and 1982, the target group was "Democrats" and "Republicans," whereas between 1978 and 2008 the target group was "Democratic Party" and "Republican Party," with the 1980 and 1982 batteries featuring both sets. In these two years, therefore, we can test whether respondents treated the parties interchangeably with party supporters, although we will not be able to clarify whether respondents were thinking of partisan voters or party leaders when providing their thermometer scores. The thermometer ratings were also part of the 2004 Issues Survey; respondents were asked to evaluate "Democrats" and "Republicans." The 2008 AP-Yahoo! News did not include feeling thermometers, but respondents rated the two parties on a four-point scale ranging from "very unfavorable" to "very favorable." In both studies, we took the difference in self-identified partisans' ratings of the out- and in-parties as a measure of net partisan affect. This variable was rescaled to range between 0 and 1. Since 1964, ANES respondents have also rated "liberals" and "conservatives" on the thermometer. We use these ratings to carry out a parallel examination of how affect toward ideological groups has fared over the years. In all the studies with the feeling thermometer questions, we use raw in- and out-group thermometer ratings as our measures of intra- and inter-party affect. However, when undertaking correlational analyses, we confine ourselves to a net difference measure (in-group score - out-group score), which is less susceptible to measurement concerns (see Winter and Berinsky 1999), while we use both raw and differenced scores for mean comparisons.6 FEELINGS CONCERNING INTER-PARTY MARRIAGE This measure of social distance was asked by Almond and Verba via the following question: "Suppose a son or daughter of yours was getting married. How would you feel if he or she married a supporter of the Republican/ Democratic (Conservative/Labor) Party? Would you be pleased, would you be displeased, or would it make no difference?" A version of this question was included in the 2008 YouGov study: "How would you feel if you had a son or daughter who married a Republican/Democrat (Conservative/Labor) supporter? Not at all upset, somewhat upset, very upset?" The same wording was adopted in the eleven-nation study, but respondents were offered a slightly altered set of options ranging from very unhappy to very happy. STEREOTYPES OF PARTY SUPPORTERS This indicator is limited to the YouGov 2008 study and, in a limited form, to the Almond and Verba study. YouGov 2008 included an extensive battery of trait 6. In all correlational analyses of the net thermometer ratings, we rescaled the difference to range between 0 and 1. 412 Iyengar, Sood, and Lelkes terms that respondents could select to describe "people who are Republicans or Democrats" (or Conservative or Labor) "supporters." The terms were patriotic, closed-minded, intelligent, hypocritical, selfish, honest, open-minded, generous, and mean. The Almond and Verba study asked respondents to think about "what sorts of people support and vote for the different parties." Respondents indicated whether a set of positive and negative terms applied to party supporters. The terms included "interested in national strength and independence," "selfish people," "intelligent people," "fascists and militarists," "betrayers of freedom," "ignorant and misguided people," and "people interested in the welfare of humanity." PARTISANSHIP All studies used in the article included either the ANES measure of party identification or a measure very similar to it. ANES has typically asked respondents about their partisanship in the following manner: "Now we would like to know something about your party preference and how you vote. Do you consider yourself a supporter of a particular political party? Which party?" Depending on the answer to this question, respondents were further asked in the ANES one of the following: "Would you call yourself a strong Republican or a not very strong Republican?" Or, "Would you call yourself a strong Democrat or a not very strong Democrat?" Or, "Do you think of yourself as closer to the Republican Party, closer to the Democratic Party, or equally close to both?" Responses to the partisanship questions were used in two different ways—to create party dummies, and as a dummy for strong partisans. Results IN- AND OUT-PARTY AFFECT IN THE UNITED STATES Our general expectation is that over time partisans' ratings of the opposing party will turn increasingly negative. Given the increasing divide between party elites and followers in their policy preferences, we might also anticipate some downturn in ratings of the in-party. However, in-party affect has changed little over time (see figure 1, which is based on table Al in the Online Appendix); Democrats and Republicans generally provide enthusiastic thermometer ratings (averaging around 70) of their own party that have remained stable over the past three decades. On the other hand, the trend in thermometer ratings of the out-party is clearly downward; on average, the rating of the out-party has dropped by some fifteen points since 1988. A similar trend appears if one just tallies the proportion of respondents giving an out-party rating of less than 50 on the thermometer scale—the proportion increases from an average of .40 in the 1980s to an average of about .53 in the 1990s, to .56 (in 2004), and .63 (in 2008). Thus, the thermometer data A Social Identity Perspective on Polarization 413 80 £70 cd co '■e cd q. ^60 cd 50 o 140 A + + cd cd i 30 20 Republican - In Party Republican - Out Party Democrat - In Party Democrat - Out Party A 1975 1980 1985 1990 1995 2000 2005 2010 Year Figure 1. Thermometer Ratings of Parties. indicate that it is the standing of the out-group that has changed; partisans like their opponents less and less.7 The latest numbers are impressive in absolute terms as well. In 2008, the average rating of the out-party was just above 30. To put this rating in some perspective, the average rating of Catholics by Protestants was 66, of "Big Business" by Democrats 51, of "gay men and lesbians (that is, homosexuals)" by Republicans 42, and of "people on welfare" by Republicans 50. One expectation is that increases in partisan affect will be concentrated among the "activists"8 (Fiorina, Abrams, and Pope 2008). And indeed, ratings 7. We observe a roughly similar pattern in the case of the "likes-dislikes" question. In 1996, 2004, and 2008, ANES respondents were asked to rate various political parties "on a scale from 0 to 10, where 0 means you strongly dislike that party and 10 means you strongly like that party." We rescored the variable to range between 0 and 1. The average level of in-party liking increased from 0.71 to 0.76, while the mean for the out-party decreased from 0.39 to 0.37. 8. Respondents who report doing three or more campaign-related activities (Fiorina, Abrams, and Pope 2008). In the ANES, respondents were asked if they had (1) attended "any political meetings, rallies, speeches, dinners, or things like that in support of a particular candidate," (2) worn "a campaign button, put a campaign sticker on [their] car, or placed a sign in [their] window or in front of [their] house," (3) done "any (other) work for one of the parties or candidates," (4) contributed money to a candidate running for public office, or (5) "talk[ed] to any people and tr[ied] to show them why they should vote for or against one of the parties or candidates?" Responses to the questions were dichotomously coded (1 = engaged in the activity, 0 = did not engage in the activity) and summed. 414 Iyengar, Sood, and Lelkes 1970 1980 1990 Year 2000 2010 Figure 2. Thermometer Ratings of Ideological Identifiers. among activists are more polarized—the difference between in-party and out-party thermometer ratings of activists and non-activists has been about 13.3 points in each survey. However, the over-time linear trend among non-activists is also significant. Moreover, it appears that the number of activists has risen sharply over the past decade—from an average of 5 percent in the preceding two decades to an average of nearly 8.5 percent in the past two presidential election cycles. This upturn in activism is likely partly attributable to increasing partisan affect that motivates partisans to participate in campaign activities. Changes in evaluations of liberals and conservatives do not exhibit the same pattern as evaluations of the parties; rather, both in-group and out-group ratings remain more or less stable over the entire series (see figure 2, based on table A2 in the Online Appendix). These data also confirm the conventional wisdom about the privileged status of conservatism in American political culture (Ellis and Stimson 2009); while Democrats' evaluations of conservatives remain in the 45-50 range over the entire period, Republicans' ratings of liberals are some 5-8 points lower. The more important point, however, is that compared to party affiliation, ideology is a weaker basis for subjective political identity (c.f. Malka and Lelkes 2010). The NES surveys also allow us to compare trends in evaluations of the parties with party supporters, albeit only for the 1980 and 1982 surveys. As shown in table 1, the differences between in-party and in-partisan ratings are minimal A Social Identity Perspective on Polarization 415 Table 1. Average Thermometer Rating of Partisans and Parties Ratings of Democrats/ Republicans Ratings of Democratic and Republican Party Difference of In-group In-party Year In-group Out-group Out-group In-party Out-party Out-party differences 1980 71.74 52.78 18.97 71.97 46.54 25.43 -6.46* 1982 72.71 49.03 23.68 73.80 44.26 29.54 -5.86* Note.—*p < .01. Tests of significance are two-tailed. All data are weighted. Source.—NES; sample limited to self-identified Republicans and Democrats only. and non-significant, while differences between out-partisan and out-party ratings amount to about 5 points (p < .01) in favor of party supporters. The average correlation between the two measures (for both parties) is strong (r = .69; p < .001), which suggests the possibility that people perhaps extend their dislike of the parties to dislike of people within the parties. The lower level of hostility directed at out-party supporters is not surprising given the evidence in psychology of a pervasive "person positivity bias"—across a variety of contexts, individuals respond more favorably when the attitude target is at the individual rather than abstract level (see Sears 1983). It is worth noting that the increased level of out-group negativity expressed toward parties appears to be unrelated to particular electoral outcomes. In 1984 and 1992, party thermometer measures were asked in both the pre- and post-election waves. When we compare the change in out-group thermometer ratings of winning and losing identifiers, there is no evidence that the "agony of defeat" has any bearing on the level of out-party sentiment. In 1984, for instance, Republican thermometer ratings of Democrats were more negative after the election than before it (p < .01), exactly the opposite of what the theory would predict. Our final piece of evidence concerning the thermometer ratings is a comparison of the thermometer ratings of political parties with ratings of other groups. The ANES has often included thermometer ratings of racial and social groups. We can thus assess changes in in-group and out-group affect associated with these alternate bases of group identity. As shown in figure 3, race and religion are weaker forms of identity than partisanship; over the past three decades, it is the partisan cleavage, rather than racial or religious divisions, that is greater. Social Distance INTER-PARTY MARRIAGE Our observations begin with the 1960 Almond-Verba five-nation survey and extend to two separate two-nation comparisons administered in 2008 and 416 Iyengar, Sood, and Lelkes at <» II I 03 £ o - Democrats/Republicans -- Democratic/Republican Party - Race (Whites/Blacks) - Religion (Catholic/Protestant) I I I i i I I i i I I i 1964 1968 1972 1976 1980 1984 19 Years i i i i i i i 1992 1996 2000 2004 2008 Figure 3. Comparing Party Polarization with Racial and Religious Polarization. 2010. The three surveys included similar questions asking how respondents would feel if their son or daughter were to enter into marriage with someone with a similar or dissimilar party affiliation. We focus on the trend over time as well as the difference in social distance between partisans in the United States and United Kingdom. As shown in figure 4 (based on table A3), we observe significantly greater increases in polarization in the American samples; by 2008, American partisans feel more apart than their British counterparts. In 1960, Almond and Verba found that only five percent of Republicans and four percent of Democrats felt "displeased" if their son or daughter married outside their political party. The number who reported feeling "pleased" by the prospect of an in-party marriage was no greater: two and three percent, respectively. Thus, the prospects of either an in- or out-party marriage within the family elicited little affective response. In 1960, the question of inter-party marriage was more consequential in the United Kingdom. Twelve percent of Conservative identifiers responded that they would be displeased by an out-party marriage. The figure was much lower for Labor supporters (three percent), but on average the level of displeasure was significantly higher in the British sample (p < .05). Thus, while party identification proved irrelevant as a marker of social distance in the A Social Identity Perspective on Polarization 417 ■c TO >. Q. Q_ i_ q_ cd 03 .c nh O c i— cc o E -6 o cd M— (/> cd CO C cd o 9- cd T3 E o -t—~ cd (f) W "O q. cd 3 L_ C CC g E ort >. c q. cd O L_ CD D. O i_ q. 4— | Republicans | Democrats Conservatives Labour c in c 1960 Proportion displeased 2008 Proportion somewhat or very upset 2010 Proportion somewhat or very unhappy Figure 4. Dissatisfaction with Inter-Party Marriage. United States, it was a cleavage with at least some inter-personal ramifications in the United Kingdom, particularly for supporters of the Conservative Party. Fifty years later, we obtained a perfect reversal of results. In the present era, it is the Americans who take their partisanship more seriously. In the 2008 YouGov survey, the proportion of Democratic and Republican identifiers who indicate feeling "somewhat upset" or "very upset" at the possibility of an out-party marriage of a child increased to 0.27 and 0.20 for Republicans and Democrats, respectively. The difference in the level of distance maintained by Republicans and Democrats was significant (p < .05). As in the case of the 1960 British results, it is the party that draws from the upper tail of the wealth distribution that exhibits greater social distance. The level of social distance in the United Kingdom increased only modestly between 1960 and 2008. Among supporters of the Conservative Party, 10 percent reported feeling somewhat or very upset; for Labor supporters, the corresponding figure was almost twice as high (19 percent). The sharp increase in social distance among Laborites may reflect the legacy of Thatcherism and the generally weakened state of the union movement in the post-Thatcher era. These developments possibly embittered Laborites, leading them to treat Tory supporters in a more exclusionary manner. Averaging across party supporters, the mean level of social distance in the United Kingdom was 0.15, considerably below the corresponding average of 0.24 in the U.S. sample (p < .001). 418 Iyengar, Sood, and Lelkes Turning to the 2010 survey, respondents were asked whether they felt somewhat or very unhappy at the prospect of inter-party marriage. Nearly half of the Republicans in the sample (0.49) selected one of these options. The corresponding level of unhappiness was 0.33 among Democrats. The difference in social distance between the parties was highly significant (p < .01). Unlike the thermometer data, where Democrats and Republicans showed roughly the same level of out-group negativity, Republicans maintain greater social distance from Democrats. As in the case of the 2008 survey, the average level of distance in the United States was considerably higher (p < .01) than in the United Kingdom. The results from both recent surveys thus stand in stark contrast to the 1960 findings.9 Party affiliation is far from an irrelevant cue when Americans think about marriage partners; significant numbers of Democrats and Republicans, but especially the latter, express displeasure at the prospect of a family member marrying into the out-party. Social distance between partisans in the United States significantly exceeds the corresponding distance in the United Kingdom. All told, the social distance indicator suggests that Americans feel much more divided along party lines today than they did fifty years ago. In the current era of polarized politics, the level of inter-party ill will is sufficient to inject partisanship into decisions that are entirely personal. Not only have Americans become more polarized over time, they now also show greater animus for the out-party than do the British, even though party identification in Britain is an expression of both political identity and class identity. TRAIT RATINGS The trait ratings data are also available for 1960 and 2008, although in very limited form for the former.10 Only two traits—intelligence and selfishness— are common to the 1960 and 2010 questionnaires, and we present results for them separately. As shown in table 2, we computed average ratings for the positive and negative traits.11 In 1960, the level of in-party bias in the positive ratings was relatively modest in both nations. Republicans gave fellow partisans the edge, but only by 0.10 on a 0-1 scale. The corresponding in-party bias was 0.12 for Democrats. In Britain, partisans were significantly more likely to 9. Two concerns potentially vitiate over-time comparisons: 1) mode differences—while the interviews in 1959-1960 were conducted face-to-face, the 2008 data come from online surveys, and 2) differences in question wording. The sheer size of the observed differences over time makes it improbable that the effect is simply an artifact of social desirability bias in face-to-face interviews, or question-wording differences. However, we acknowledge that the real effect may be somewhat smaller. 10. The Almond-Verba survey included two positive traits—intelligent and interested in the welfare of humanity—and four negative traits—selfish, ignorant, betrayers of freedom, and fascist. 11. For the complete set of trait ratings, see table A5 in the appendix. Table 2. Mean Ratings of Selected Traits, 1960,2008 United States United Kingdom Difference (U.S. - UK) Year Items In-party Out-party In - Out In-party Out-party In - Out In-party Out-party In - Out 1960 Selected traits" Intelligent .33 .27 .06* .26 .09 17*** _ 07** _17*** _11*** Selfish .04 .21 _17*** .02 .29 _ 27*** -.02* 0g*** 10*** Average number Positive traits .47 .35 12*** .34 .12 22*** 13*** 23*** _10*** Negative traits .01 .08 _07*** .01 .13 _12*** .00 -.05*** .05*** Positive - Negative .46 .27 19*** .33 -.01 34*** 13*** 3g*** _15*** N 669 759 2008 Selected Traitsa Intelligent .62 .14 48*** .38 .10 29*** _ 24*** -.04* 20*** Selfish .04 .47 _ 42*** .03 .28 _ 25*** -.01 _19*** _18*** Closed-minded .08 .62 — 53*** .05 .47 _ 42*** -.03** _14*** _11*** Average number Positive traits 2.86 .62 2 24*** 1.93 .42 \ 52*** 93*** 20*** 72*** Negative traits .22 1.93 —1 70*** .13 1.29 -1.15*** .09** g4*** -.55*** Positive - Negative -1.31 2.63 3 94*** -.87 1.80 2.67*** _ 44*** 84*** I 27*** N 656 817 Note.—*p < .05; **p < .01; ***p < .001. Tests of significance are two-tailed. All data are weighted. Sample limited to Republicans and Democrats in the United States, and to Labor and Conservative supporters in the UK. Source.—Almond and Verba in 1960 and YouGov/Polimetrix in 2008. aFor a full list of trait ratings, see the Online Appendix. S10Z '£ itremref uo jsrux SHN sjBjxdsoji auXx uodn gpsEOMgjsj ys /§jo's^ujnofpjojxo'bod//:djjquiojj popeojuMOQ 420 Iyengar, Sood, and Lelkes offer more positive ratings of their party; for Conservatives, the in-party bias was 0.25; for Labor, it was 0.18. The difference in the average level of in-party bias across the two nations (0.12 in the United States and 0.22 in Britain) was significant (p < .01). Negative stereotypes of party supporters revealed the very same pattern— relatively lower levels of out-party stereotyping among American partisans. In 1960, Democrats and Republicans were more apt to impute negative traits to their opponents by a margin of less than 0.1. (Note, however, that many more were willing to label out-partisans as selfish.) Partisans in the UK were only slightly more negative in their beliefs about the opposition. The cross-national difference in out-group bias, although modest, was significant (difference = .05, p< .01). Between 1960 and 2008, stereotyping of partisan supporters and opponents increased exponentially. Among the Americans, in-party bias—the tendency to view one's party more favorably than the opposition—climbed from 0.47 to 2.86!12 The increase in negative stereotyping of the out-party was just as extensive; in comparison with 1960, Democrats and Republicans were nearly fifty percent more likely to associate negative traits with opponents than supporters in 2010. When we limit the over-time comparisons to the pair of common indicators, the trend is no less spectacular. The level of in-group favoritism on intelligence increased from 0.06 to 0.48 in the U.S. sample and from 0.17 to 0.29 in the UK sample. The tendency to rate opponents more than supporters as selfish increased from 0.21 to 0.47 in the American samples and actually declined from 0.29 to 0.28 in the UK data. The dramatic strengthening of party stereotypes in the United States was not matched in Britain. In the case of positive traits, in-party bias increased to 0.41 among Conservative identifiers (an increase of 200 percent) and remained stable at 0.22 among Labor supporters. The accentuation in negative beliefs about the other side increased to 0.3 for Conservatives and 0.28 for Laborites. Thus, by 2008, party stereotypes were significantly more polarized among Americans.13 The changes over time and the pattern of cross-national differences in the trait ratings both indicate that party stereotypes were initially weak, but 12. The 2008 survey included five positive traits—generous, honest, intelligent, open-minded, and patriotic. The negative set consisted of mean, hypocritical, selfish, and closed-minded. Once again, the over-time comparisons may be compromised by the use of different trait terms. However, as discussed below, the results for the two traits asked in all the surveys (intelligent and selfish) also show accelerated polarization in the United States. 13. Trait ratings of partisans can be thought to represent an underlying "latent trait" of partisan affect. The more the respondent dislikes the out-party, the more likely the respondent is to ascribe a negative trait to out-party supporters, and a positive trait to in-party supporters. Fitting an IRT model to the trait ratings replicates the pattern based on mean comparisons—Americans are considerably more polarized than their counterparts across the Atlantic. (These results are available from the authors.) A Social Identity Perspective on Polarization 421 somewhat more frequent in the United Kingdom. By 2008, however, stereotyping was widespread, and it was the Americans who expressed more positive views of party supporters and negative views of their opponents. Along with the inter-marriage question, these results indicate that partisan identity is stronger among American partisans. In summary, our three indicators of affective polarization all demonstrate that partisans in America are increasingly divided. The sense of partisan identity is increasingly associated with a Manichean, "us against them" view of the political world. Democrats and Republicans harbor generally negative feelings toward their opponents. Stereotypes of party supporters have become increasingly differentiated; positive traits accrue to members of the in-party, while negative traits are ascribed to opponents. There is sufficient animosity to make partisan affiliation relevant to inter-personal relations. Today, American partisans are highly polarized in their feelings about each other. Possible Explanations We consider two possible explanations for affective polarization—movement in policy attitudes among the masses and the elites, and exposure to increasingly negative political campaigns. We analyze potential spillover from ideological to affective polarization by examining the impact of policy preferences on thermometer ratings of the parties. We also report two tests of the relationship between exposure to political campaigns and polarization. SPILLOVER BETWEEN IDEOLOGICAL AND AFFECTIVE POLARIZATION Is affective polarization simply a symptom of divergent movement in policy attitudes among both partisan supporters and party elites? The evidence suggests otherwise. First, if ideological disagreement has contributed to affective polarization, we should observe considerably sharper thermometer differences among sorted partisans, i.e., liberal Democrats and conservative Republicans. However, the differences in thermometer ratings among the sorted partisans are only modestly higher (see table A4 in the Online Appendix). As a more precise test of possible spillover between ideological and affective polarization we used the 1988 and 2004 ANES data sets to examine the impact of policy preferences on partisan affect, and to assess whether this relationship has changed over time.14 Using Layman and Carsey (2002) as our guide, we assessed preferences on the two major issue dimensions in contemporary American politics—social welfare and cultural issues. Preferences on the 14. The 1988 ANES study was used, as it was the first to include all three cultural attitude issues. We use the 2004 data because the inclusion of innumerable question-wording experiments on the issue preference scales in the 2008 ANES study makes any analysis of issue-based polarization daunting. 422 Iyengar, Sood, and Lelkes former were measured via items soliciting attitudes toward government spending on social security, support for a government health insurance plan and for more government services, and the expectation that government should guarantee jobs and a standard of living. Cultural preferences were measured via items asking respondents about their attitudes toward abortion, laws, gay rights, and gender equality.15 We imputed missing data using Amelia II (Honaker, King, and Blackwell 2006) and created 25 data sets. Next, using confirmatory factor analysis, we computed factor scores from a two-factor (Social Welfare and Culture) solution for each imputed data set. These scores were scaled so that higher values indicated more conservative (anti-spending, pro-traditional values) stances for Republicans and more liberal stances for Democrats. To test the relationship between policy attitudes and partisan affect, we regressed the differenced in-party and out-party thermometer scores on the factor scores corresponding to each policy dimension16 as well as a number of other variables associated with partisan affect, including strength of partisan identity, interest in elections, and a set of demographic factors.17 All variables were coded to lie between 0 and 1. A /7-value of less than .05 indicated that the 95-percent confidence interval did not include zero; because we utilized imputed data, confidence intervals were calculated using Rubin's degrees-of-freedom estimate (Rubin 1987). The results are presented in table 3. Surprisingly, given the "heat" associated with cultural issues, it was the social welfare dimension that more strongly predicted inter-party animus.18 Social welfare preferences significantly influenced partisan affect (b = .20, p < .05 for Democrats in 1988 and 2004; b = .12, p < .05 and b = .18, p < .05 for Republicans in 1988 and 2004, respectively),19 while cultural preferences exerted no influence at all.20 The 12-20-point difference in thermometer 15. The only variable that appeared in Layman and Carsey's analysis but not our own dealt with attitudes toward food-stamp spending. We excluded this variable as it did not appear in the 2004 election study. 16. Other plausible specifications include regressing difference in thermometer ratings on difference in self-placement (or in-party placement) and out-party placement. Our specification is defensible for the following two reasons: (1) party placements on issues are likely endogenous to partisan affect (though so are, to likely a far lesser degree, self-placements), and (2) following the first point, given the plausible scenario of out-party placement being negatively correlated to self-placement, our specification gives the more conservative estimate of the effects of differences in self-placement and out-party placement on partisan affect. 17. We split our analyses by party for clarity and to account for possible differences across parties in the relationship between ideological positions and partisan affect. 18. The relative influence of economic over cultural policy preferences as predictors of party affect is consistent with Bartels' findings (Bartels 2010, p. 86). 19. Between 1988 and 2004, the effect of economic attitudes on affective polarization did not change among Democrats (£(1,544) = 0.76, n.s.), but was strengthened among Republicans (f(l,315)= 5.16,/?<.001). 20. Taking out the strength of partisan identification predictor (which likely tracks both social identity and policy preferences) increases the size of the coefficients on policy preferences, but the improvement is at least in part a result of omitting an indicator of social identity. A Social Identity Perspective on Polarization 423 Table 3. Effect of Policy Preferences on Net Partisan Affect, OLS Coefficients (standard errors in parentheses) 1988 2004 Predictors Democrats Republicans Democrats Republicans Intercept .23* .04 .19* .00 (.05) (.05) (.08) (.09) Cultural attitudes -.06 .04 .05 .04 (.04) (.04) (.07) (.06) Economic attitudes .20* .12* .19* .19* (.05) (.05) (.07) (.06) Strong identifier .18* .17* .26* .22* (.02) (.02) (.02) (.02) Political interest .05 .06* .07* .12* (.02) (.03) (.03) (.03) Gender: female .01 -.02 .04 .02 (.02) (.02) (.02) (.02) Race: White -.04 .11* .00 .01 (.02) (.04) (.02) (.03) Region: South -.03 .06* .05* .05* (.02) (.02) (.02) (.02) Education: high school -.05* .00 -.16* .05 (.02) (.03) (.06) (.08) Education: some college -.02 .00 -.16* .03 (.03) (.03) (.06) (.08) Education: college or higher -.09* -.02 -.16* .02 (.04) (.04) (.07) (.08) N 830 954 592 485 Adjusted 7?-squared .16 .17 .25 .27 Note.—* indicates that the 95-percent confidence interval does not contain zero. Confidence intervals were calculated using Rubin's (1987) multiple-imputation degrees-of-freedom estimate. All variables ranged between 0 and 1. Larger values on the cultural attitudes and economic attitudes variables indicate more liberal attitudes among Democrats and more conservative attitudes among Republicans. scores between the least and most ideological partisans was equivalent to the difference in the in-/out-party thermometer differences between weak and strong party identifiers (b = .18, p < .05 and b = .26, p < .05 for Democrats in 1988 and 2004; b = .ll,p< .05 and b = .22, p < .05 for Republicans). The other thing to note is that coefficients for strong partisan identification are considerably stronger in 2004. This suggests a rising premium for partisan identification, beyond issue preferences. Finally, when we regress in-party thermometer scores on the same set of variables, the coefficients on policy preferences are uniformly weaker (and outside of Social Welfare preferences 424 Iyengar, Sood, and Lelkes among Democrats in 1988, always insignificant) than when we regress out-party thermometer scores on policy preferences. In sum, the evidence presented above—the non-effects of sorting on the level of affective polarization, and the moderate to weak effects of policy preferences on net partisan affect—is at odds with the view that partisan affect is driven primarily by ideological affinity. Instead, the data suggest that the two indicators tap distinct forms of polarization. The weak association between ideological and affective polarization is consistent with the vast literature demonstrating that partisan identity in the United States has only weak ideological underpinnings. For parties to be seen in ideological terms would require an awareness of their policy stances, yet most Americans have difficulty correctly locating parties on issue scales (see, e.g., Delli Carpini and Keeter 1993). As we emphasize in the concluding section, partisan identities are primarily affective attachments. EXPOSURE TO NEGATIVE CAMPAIGNS Next, we analyze the degree to which partisan affect can be attributed to increasingly loud and negative political campaigns. Our identification strategy rests upon exploiting exogenous variation in actual campaigning based on electoral college incentives, possibility of winning a state based on early polls and punditry, etc. These strategic considerations mean that otherwise identical voters are exposed to vastly different amounts of campaigning depending on the area they live in. "Battleground states"—states that see particularly intense campaigning—are a prominent example of the intersection between geography and campaign strategy. We naturally expect partisans living in these states to have a stronger partisan affect than those living in non-battleground states. We begin with data from the 2004 Blair Center Election Study.21 To estimate the effects of residence in a battleground state22 on net partisan affect, we regressed the difference between the in- and out-party thermometer ratings on battleground status, political interest, partisan identity, strength of partisan identification, gender, age, race, and education.23 As displayed in the first 21. The decision to use this survey instead of the ANES was based on sample size. After excluding independents, the ANES sample was substantially smaller (N = 1,066 in the ANES versus N= 2,276 in the Blair Survey). 22. States with the two parties' vote difference of less than 5 percent in the 2000 elections were defined as battleground states: FL, IA, MO, MN, NH, NM, NV, OH, OR, PA, and WI (Hillygus and Shields 2005). Ideally, we would like to use battleground-state residence as an instrument for exposure and estimate the impact of exposure (Ashworth and Clinton 2007). Here, we opt for this reduced model due to the lack of ad recall measures in the survey data. Thus, our estimate is of the impact of difference in all campaign-related activity between battleground and non-battleground states. 23. The party favorability, political interest, and partisanship questions used in this survey were parallel in form to those in the ANES; age was measured in years, race was entered as a dummy for non-Hispanic whites, and education was measured using four categories (less than HS, HS, some college, college or higher). A Social Identity Perspective on Polarization 425 Table 4. Effect of Negative Advertising on Net Partisan Affect in 2004, OLS Coefficients In-party minus out-party thermometer ratings Predictors Model 1 Model 2a Battleground-state residence .02* — Total negative ads run in state during general elections - .19* Total (other) ads run in state - -.11* Republican 04*** 04*** Strong identifier 20*** 2\*** Political interest .03* .03* Age -.05** -.05** Female .01+ .02** Race: white — 03*** -.03** Education: high school .00 .00 Education: some college _.04** -.03* Education: college or higher -.05*** -.05*** Adjusted 7?-squared .27 .28 N 2,276 1,882 Note.—vp < .10; *p < .05; **p < .01; ***p < .001. Tests of significance are two-tailed. Intercept not reported. W for Model 2 is lower because respondents living in states not covered fully by the advertising data were excluded. column of table 4, battleground-state residence positively predicted net partisan affect (b = .02, p < .05). As a robustness test, we combined the number of general election attack ads24 run in each state (as a lower-level geographic indicator was not available) and used this measure of the total volume of negative advertising to predict partisan affect, controlling for other ads run in the state along with all the other covariates used in the previous analysis. As shown in the second column of table 4, the volume of general election attack ads aired in a state25 positively predicted net partisan affect (b = .19, p < .05).26 We acknowledge that the effect size associated with exposure to negative ads is modest. This is only to be expected, since advertising contributes only a 24. We pooled presidential and gubernatorial, Senate, House advertisement data. As is common practice, both "attack" and "contrast" ads were coded as "negative." 25. Both volume of general election attack ads and volume of other ads were jointly rescaled to lie on the same scale, ranging from 0 to 1. A unit increase corresponds to an increase of about 130,000 ads. 26. Taking a log of number of ads run (both attack ads and other ads) leaves the substantive results unchanged. 426 Iyengar, Sood, and Lelkes Table 5. Effect of the 2008 Campaign on Net Partisan Affect, Hierarchical Linear Model Coefficients Predictors In-party minus out-party favorability ratings Battleground -.01 Wave .02* Wave x Battleground .02* Republican .06* Strong identifier .36* Age .00 Female .00 Race: white -.00 Education: high school .02 Education: some college .02 Education: college or higher .01 Log likelihood 6,669 N 1,401 Note.—Tests of significance are two-tailed. Intercept, random effects for DMA, and individual respondents not reported; * indicates that the 95-percent HPD interval does not contain zero. small share of the total information stream in presidential campaigns and voters are more likely to encounter campaign messages through news coverage. For instance, while campaigns undoubtedly air more ads in battleground states, the ads are recycled in the national media, thus exposing people across the country. Our second test of the effects of campaigns on affective polarization takes advantage of the panel structure of the 2008 AP-Yahoo! News study to examine changes over time in the level of polarization in relation to campaign exposure. We have twin predictions—a general trend toward increased affective polarization as the campaign progresses, with a steeper temporal gradient in the battleground states. We implemented a hierarchical linear model for predicting net partisan affect from battleground-state residence, time of interview, gender, race, residence in the South, and education, with random effects for each respondent, and designated market area.27 As shown in table 5, affective polarization increased significantly over the course of the campaign (b = .02, p < .05). Moreover, as predicted, partisans in battleground states were especially likely to become more polarized. In fact, the interaction between time and battleground-state 27. Favorability of parties was measured using a four-point scale running from "Very unfavorable" (0) to "Very favorable" (1), age was measured in years, education was coded the same way as in previous analyses, the political interest variable was used from the November 2007 wave to avoid potential post-treatment bias, and the wave variable ran from 1 to 9 but was rescaled to lie between 0 and 1. The following states were defined as battleground states: CO, FL, IA, MI, MN, NC, NM, NH, NV, OH, OR, PA, WI, and WV (see Jackman and Vavreck 2009). A Social Identity Perspective on Polarization 427 residence indicates that the effects of the campaign doubled among residents of battleground states. Overall, our analyses suggest that greater levels of negativity in advertising campaigns and general exposure to political campaigns both contribute to higher levels of affective polarization. Campaigns reinforce voters' sense of partisan identity and confirm stereotypical beliefs about supporters and opponents (for a similar argument concerning identification with the nation state in the aftermath of national security crises, see Althaus and Coe 2011). These results implicate the "son et lumiere" of American politics as a possible cause of increased out-group animus. Discussion Is the American public politically polarized? The answer depends on the indicator. On construct validity grounds, we contend that affect is a more appropriate indicator of mass polarization than ideology. Beginning with the pioneering work of Lane (1959) and Converse (1964), political scientists have repeatedly demonstrated that the vast majority of the public does not think about parties in ideological terms and that their ties to the political world are instead affective, based on a primordial sense of partisan identity that is acquired very early in life and persists over the entire life cycle (Alwin, Cohen, and Newcomb 1991, p. 159; Green, Palmquist, and Schickler 2002; Hyman 1959; Sears and Funk 1999; Jennings, Stoker, and Bowers 2009). As our examination of the interplay between policy preferences and party thermometer ratings reveals, "principled" dislike of the out-party makes up only a small component of inter-party affect. As we have suggested, the more plausible explanation of intensified inter-party animus lies in the rhetoric of political campaigns. Virtually every study of campaign advertising documents the steep increase in the frequency of attacks and counterattacks (Benoit 2001; Geer 2010). The tendency of the media to recycle the candidates' negative messages only confirms partisans' suspicions about those on the other side. Exposure to loud negative campaigns is very likely not the strongest factor, much less the only factor, contributing to affective polarization. Technology has facilitated citizens' ability to seek out information sources they find agreeable and tune out others that prove dissonant (see Iyengar and Hahn 2009; Stroud 2008, 2010). As consumers begin to exercise their ability to select "friendly" sources, an increasing number of news providers deliver slanted news (Mullainathan and Shleifer 2005; Gentzkow and Shapiro 2006). As partisan news sources expand their share of the market, the congruence between prior beliefs and incoming information will only increase.28 The ability to 28. Such self-selection is also likely to incentivize elite movement toward the poles. See also Gul and Pesendorfer (2011). 428 Iyengar, Sood, and Lelkes select information sources that routinely denigrate the out-party is likely to lead to increased out-party animus (Levendusky, forthcoming). Naturally, the increased level of affective polarization poses considerable challenges to the democratic process. Partisan bias in perceptions of economic conditions means that voters will fail to credit opposing-party incumbents when the economy grows under their stewardship and fail to penalize in-party incumbents whose economic performance is suspect. Biased beliefs about opposing elites—that they are duplicitous, self-interested, stupid, etc.—make it improbable that elites can persuade out-party partisans. A more serious concern is that those who impugn the motives and character of political opponents are less likely to treat as legitimate the decisions and policies enacted when the opponents control government, and may also be less satisfied with institutions that respond to popular will. A measure of "satisfaction with democracy" shows an increasing divide between partisans since 2000.29 In recent years, partisans on the losing side have been substantially more dissatisfied. Dissatisfaction with policy outcomes and democratic institutions can escalate into mass protest, and in some cases to acts of violence. In closing, we would like to reiterate that our principal goal has been to reframe the debate over polarization in terms of affect rather than ideology. The evidence is strong that partisans are affectively polarized. It is less clear what exactly underlies this development. The evidence we have presented suggests that political campaigns are implicated, and there are reasons to believe that the fragmentation of the media market is a further contributory factor, but future research will need to address these possible explanations more systematically. Supplementary Data Supplementary data are freely available online at http://poq.oxfordjournals.org/. References Abramowitz, Alan. 2010. The Disappearing Center: Engaged Citizens, Polarization, and American Democracy. New Haven, CT: Yale University Press. Abramowitz, Alan I., and Kyle L. Saunders. 2008. "Is Polarization a Myth?" Journal of Politics 70:542-55. Aldrich, John H. 1995. Why Parties? Chicago: University of Chicago Press. Almond, Gabriel A., and Sydney Verba. 1960. Civic Culture Study, 1959-1960 [Computer file]. ICPSR07201-v2. Ann Arbor, MI: Inter-University Consortium for Political and Social Research [distributor], 2009-02-12. 29. In the ANES, respondents were asked "on the whole, are you very satisfied, fairly satisfied, not very satisfied, or not at all satisfied with the way democracy works in the United States?" We created a dichotomous coding comparing those who were "very" and "fairly" satisfied with the rest. A Social Identity Perspective on Polarization 429 Alwin, Duane R, Ronald L. Cohen, and Theodore M. Newcomb. 1991. Political Attitudes over the Life Span: The Bennington Women after Fifty Years. Madison: University of Wisconsin Press. Althaus, Scott L., and Kevin Coe. 2011. "Priming Patriots: Social Identity Processes and the Dynamics of Public Support for War." Public Opinion Quarterly 75(l):65-85. Ashworth, Scott, and Joshua D. Clinton. 2007. "Does Advertising Exposure Affect Turnout?" Quarterly Journal of Political Science 2(1):27—41. Bartels, Larry M. 2010. Unequal Democracy: The Political Economy of the New Gilded Age. Princeton, NJ: Princeton University Press. Benoit, William L. 2001. "The Functional Approach to Presidential Television Spots: Acclaiming, Attacking, and Defending 1952-2000." Communication Studies 52(2): 109-26. Billig, M. G., and Henri Tajfel. 1973. "Social Categorization and Similarity in Intergroup Behavior." European Journal of Social Psychology 3(l):27-52. Bogardus, Emory S. 1947. "Measurement of Personal-Group Relations." Sociometry 10(4): 306-311. Burden, Barry C. 2001. "The Polarizing Effects of Congressional Primaries." In Congressional Primaries and the Politics of Representation, edited by Peter F. Galderisi, Marni Ezra, and Michael Lyons, 95-115. Lanham, MD: Rowman & Littlefield. Campbell, James E. 2008. "Presidential Politics in a Polarized Nation: The Reelection of George W. Bush." In The George W. Bush Legacy, edited by Colin Campbell, Bert A. Rockman, and Andrew Rudalevige, 21-44. Washington, DC: CQ Press. Converse, Philip E. 1964. "The Nature of Belief Systems in Mass Publics." In Ideology and Discontent, edited by David Apter, 206-61. New York: Free Press. Delli Carpini, Michael X., and Scott Keeter. 1993. "Measuring Political Knowledge: Putting First Things First." American Journal of Political Science 37(4): 1179-1206. Ellis, Christopher, and James A. Stimson. 2009. "Symbolic Ideology in the American Electorate." Electoral Studies 28(2):388-402. Fiorina, Morris P., and Samuel J. Abrams. 2009. Disconnect: The Breakdown of Representation in American Politics. Norman: University of Oklahoma Press. Fiorina, Morris, Samuel Abrams, and Jeremy Pope. 2005. Culture War? The Myth of a Polarized America. New York: Pearson Longman. -. 2008. "Polarization in the American Public: Misconceptions and Misreadings." Journal of Politics 70:556-60. Fleisher, Richard, and Jon R. Bond. 2001. "Evidence of Increasing Polarization Among Ordinary Citizens." In American Political Parties: Decline or Resurgence?, edited by Jeffrey E. Cohen, Richard Fleisher, and Paul Kantor, 55-77. Washington, DC: CQ Press. Gaertner, Samuel L., John F. Dovidio, Phyllis A. Anastasio, Betty A. Bachman, and Mary C. Rust. 1993. "The Common In-Group Identity Model: Recategorization and the Reduction of Intergroup Bias." European Review of Social Psychology 4:1-26. Geer, John G. 2010. Fanning the Flames: The News Media's Role in the Rise of Negativity in Presidential Campaigns. Working Paper. Available at http://www.hks.harvard.edu/presspol/ publications/papers/discussion_papers/d55_geer.pdf. Gentzkow, Matthew, and Jesse M. Shapiro. 2006. "Media Bias and Reputation." Journal of Political Economy 114(2):280-316. Goldstein, Kenneth, and Joel Rivlin. 2007a. "Congressional and Gubernatorial Advertising, 2003- 2004." Combined File [data set]. Final release. Madison: University of Wisconsin Advertising Project, Department of Political Science at the University of Wisconsin-Madison. -. 2007b. "Presidential Advertising, 2003-2004." Combined File [data set]. Final release. Madison, WI: University of Wisconsin Advertising Project, Department of Political Science at the University of Wisconsin-Madison. Green, Donald P., Bradley Palmquist, and Eric Schickler. 2002. Partisan Hearts and Minds: Political Parties and the Social Identities of Voters. New Haven, CT: Yale University Press. Gul, Faruk, and Wolfgang Pesendorfer. 2011. "Media and Policy." Working Paper. 430 Iyengar, Sood, and Lelkes Hetherington, Marc. 2002. "Resurgent Mass Partisanship: The Role of Elite Polarization." American Political Science Review 95(3):619-31. -. 2009. "Putting Polarization in Perspective." British Journal of Political Science 39:413^-8. Hetherington, Marc J., and Jonathan D. Weiler. 2009. Authoritarianism and Polarization in American Politics. Cambridge, UK: Cambridge University Press. Hillygus, D. Sunshine, and Todd G. Shields. 2004. Blair Center Election Study, http://hdl.handle.net/ 1902.1/10743. UNF:3:hs9t6ruiaBI12UjfkD05pQ== VI. Hillygus, D. Sunshine, and Todd G. Shields. 2005. "Moral Issues and Voter Decision Making in the 2004 Presidential Election." PS: Political Science and Politics 38(2):225-244. Honaker, James, Gary King, and Matthew Blackwell. 2006. Amelia II: A Program for Missing Data. http://gking.harvard.edu/amelia. Hyman, Herbert H. 1959. Political Socialization. Glencoe, IL: Free Press. Iyengar, Shanto, and Kyu S. Hahn. 2009. "Red Media, Blue Media: Evidence of Ideological Selectivity in Media Use." Journal of Communication 59(1): 19-39. Jackman, Simon, and Lynn Vavreck. 2009. "The Magic of the Battleground: Uncertainty, Learning, and Changing Information Environments in the 2008 Presidential Campaign." Paper presented at the Annual Meeting of the Midwest Political Science Association, Chicago. Jacobson, Gary C. 2000. The Politics of Congressional Elections. New York: Longman. Jennings, M. Kent, L. Stoker, and Jake Bowers. 2009. "Politics across Generations: Family Transmission Reexamined." Journal of Politics 71(3):782-99. Lane, Robert E. 1959. "Fathers and Sons: Foundations of Political Belief." American Sociological Review 24:502-11. Layman, Geoffrey C. 1999. '"Culture Wars' in the American Party System." American Politics Research 27:89-121. Layman, Geoffrey C, and Thomas Carsey. 2002. "Party Polarization and 'Conflict Extension' in the American Electorate." American Journal of Political Science 46:786-802. Layman, Geoffrey C, Thomas M. Carsey, and J. M. Horowitz. 2006. "Party Polarization in American Politics: Characteristics, Causes, and Consequences." Annual Review of Political Science 9:83-110. Levendusky, Matthew. 2009. "The Microfoundations of Mass Polarization." Political Analysis 17:162-76. -. Forthcoming. "Partisan Media Exposure and Attitudes Towards the Opposition." Political Communication. Malka, Ariel, and Yphtach Lelkes. 2010. "More Than Ideology: Conservative-Liberal Identity and Receptivity to Political Cues." Social Justice Research 23:156-88. McCarty, Nolan, Keith Poole, and Howard Rosenthal. 2001. "The Hunt for Party Discipline in Congress." American Political Science Review 95:673-88. -. 2006. Polarized America: The Dance of Ideology and Unequal Riches. Cambridge, MA: MIT Press. Mullainathan, Sendhil, and Andrei Shleifer. 2005. "The Market for News." American Economic Review 95:1031-53. Oakes, Penelope J. 1987. "The Salience of Social Categories." In Rediscovering the Social Group, edited by John C. Turner, Michael A. Hogg, Penelope J. Oakes, Steven D. Reicher, and Margaret S. Wetherell, 117^1. Oxford: Basil Blackwell. Poole, Keith T, Howard Rosenthal, and Kenneth Koford. 1991. "On Dimensionalizing Roll Call Votes in the U.S. Congress." American Political Science Review 85:955-75. Richardson, Bradley M. 1991. "European Party Loyalties Revisited." American Political Science Review 85:751-775. Rivers, Douglas, and Delia Bailey. 2009. "Inference from matched samples in the 2008 U.S. national elections." Proceedings of the Joint Statistical Meetings, 627-639. Rohde, David W. 1991. Parties and Leaders in the Post-Reform House. Chicago: University of Chicago Press. Rubin, Donald B. 1987. Multiple Imputation for Nonresponse in Surveys. New York: Wiley. A Social Identity Perspective on Polarization 431 Schlozman, Kay L., and John T. Tierney. 1986. Organized Interests and American Democracy. New York: Harper & Row. Sears, David O. 1983. "The Person-Positivity Bias." Journal of Personality and Social Psychology 44(l):233-50. Sears, David O., and Carolyn L. Funk. 1999. "Evidence of Long-Term Persistence of Adults' Political Predispositions." Journal of Politics 61:1-28. Stonecash, Jeffrey M., Mark D. Brewer, and Mack D. Mariani. 2003. Diverging Parties: Social Change, Realignment, and Party Polarization. Boulder, CO: Westview. Stroud, Natalie J. 2008. "Media Use and Political Predispositions: Revisiting the Concept of Selective Exposure." Political Behavior 30(3):341-66. -. 2010. "Polarization and Partisan Selective Exposure." Journal of Communication 60:556-76. Tajfel, H. 1970. "Experiments in Intergroup Discrimination." Scientific American 223(5):96-102. Tajfel, Henri, M. G. Billig, R. P. Bundy, and Claude Flament. 1971. "Social Categorization and Intergroup Behavior." European Journal of Social Psychology 1(2): 149-78. Tajfel, Henri, and John C. Turner. 1979. "An Integrative Theory of Intergroup Conflict." In The Social Psychology of Intergroup Relations, edited by W. G. Austin and Stephen Worchel, 33- 47. Monterey, CA: Brooks/Cole. Thompson, Hunter S. 1973. Fear and Loathing: On the Campaign Trail '72. New York: Straight Arrow Books. Winter, Nicholas, and Adam J. Berinsky. 1999. "What's Your Temperature? Thermometer Ratings and Political Analysis." Paper presented at the Annual Meeting of the American Political Science Association, Atlanta.