Affective Polarization or Partisan Disdain? Untangling a Dislike for the Opposing Party from a Dislike of Partisanship

Transcription

1 Affective Polarization or Partisan Disdain? Untangling a Dislike for the Opposing Party from a Dislike of Partisanship Conditionally Accepted at Public Opinion Quarterly Samara Klar University of Arizona School of Government and Public Policy 315 Social Sciences Building 1145 E. South Campus Drive Tucson, AZ klar@ .arizona.edu Yanna Krupnikov Stony Brook University Department of Political Science S723 Social and Behavioral Sciences Stony Brook, NY yanna.krupnikov@stonybrook.edu John Barry Ryan (Corresponding Author) Stony Brook University Department of Political Science N711 Social and Behavioral Sciences Stony Brook, NY john.ryan@stonybrook.edu Running Header: Affective Polarization or Partisan Disdain? Word Count: 2746

2 Author Note SAMARA KLAR is an assistant professor in the School of Government and Public Policy at the University of Arizona, Tucson, AZ, USA. YANNA KRUPNIKOV and JOHN BARRY RYAN are associate professors of Political Science at Stony Brook, University, Stony Brook, NY, USA. Funding This research was conducted with a grant from Time-Sharing Experiments in the Social Sciences. Corresponding Author Contact Information Address correspondence to John Barry Ryan, Stony Brook University, Political Science Department, N711 Social and Behavioral Sciences, Stony Brook, NY 11794, USA; john.ryan@stonybrook.edu. 1

3 Abstract Recent scholarship suggests that American partisans dislike other party members so much that partisanship has become the main social divide in modern politics. We argue that at least one measure of this affective polarization conflates a dislike for members of the other party with a dislike for partisanship in general. The measure asks people how they feel about their child marrying someone from another party. What seems like negative affect toward the other party is, in fact, negative affect toward partisans from either side of the aisle and political discussion in general. Relying on two national experiments, we demonstrate that, although some Americans are politically polarized, more simply want to avoid talking about politics. In fact, many people do not want their child to marry someone from their own party if that hypothetical in-law were to discuss politics frequently. In supplementary analyses, we use ANES feeling thermometers to show that in-party feeling thermometer ratings have decreased in recent years among weak and leaning partisans. As a result, the feeling thermometer results confirm the conclusion from the experiments. Polarization is a phenomenon concentrated in the one-third of Americans who consider themselves strong partisans. More individuals are averse to partisan politics. The analyses demonstrate how affective polarization exists alongside weakening partisan identities. 2

4 Contemporary scholars and journalists show great interest in the growing partisan divide among Americans. Although scholars debate whether this divide is based on actual differences in issue preferences or merely on perceived differences (Mason 2015, Levendusky and Malhotra 2016), most acknowledge that affective polarization view[ing] opposing partisans negatively and copartisans positively (Iyengar and Westwood 2015, p. 691) appears to be increasing. In this study, however, we find that at least one widely cited measure of affective polarization overstates the amount of affective polarization. 1 The measure asks partisans how they would feel if their child were to marry someone from the other political party. As many as half of respondents report that their child marrying someone from the other party would make them unhappy, indicating, according to Iyengar et al (2012), a large amount of affective polarization. Among the potential issues with this measure, the most problematic is that it conflates two distinct phenomena: (1) affective polarization; and (2) a dislike for political parties generally (Klar and Krupnikov 2016). Further, the measure is included alongside other political questions, heightening the salience of partisan considerations. Finally, respondents only hear about the hypothetical in-law s partisanship, but not other traits or descriptors, implying partisanship is a particularly important aspect of the potential inlaw s identity. 1 As of September 2017, Iyengar et al. (2012), which includes the measure, has been cited between 150 (Scopus) and 408 (Google Scholar). Not all citations are to the measure. 3

5 To examine the effects of these features on measured affective polarization, we conduct a survey experiment on two nationally representative samples at two points in time during 2016, with some respondents interviewed in both samples. The experiment separates a dislike for parties in general from a specific dislike for the out-party by asking respondents how they would feel about a child marrying an individual from both parties. Further, the experiment is part of a survey that does not ask other political questions. Finally, we tell respondents how frequently the hypothetical in-law discusses politics, thereby providing context about how important partisan politics is to hypothetical in-law. We find that affective polarization is a dominant trait only among the one-third of respondents who identify as strong partisans and they become more polarized during the 2016 presidential campaign. The majority of individuals are not affectively polarized -- rather many are averse to partisan politics. Because this question is only one possible measure of affective polarization, in the appendix, we use ANES feeling thermometers to confirm affective polarization is largely confined to strong partisans. Argument We suggest that studies measuring affective polarization inadvertently measure two distinct concepts: (1) dislike for the out-party; and (2) dislike for partisanship in general. As affective polarization has ostensibly grown, the percentage of people reporting that they dislike both parties has also increased (Smith 2015). Moreover, both a dislike of partisanship and a genuine dislike of the out-party manifest themselves in lower ratings of the out-party because people are more willing to publicly denigrate the out-party even if they actually dislike both parties (Groenendyk 2013). 4

6 Of primary interest in this paper is a measure of affective polarization based on the Social Distance Scale, originally developed by Bogardus (1926) to measure social distance between racial and ethnic groups. The original scale included seven items with willingness to marry a member of a particular group indicating the least social distance. Almond and Verba (1963) adapted the scale to political partisanship asking respondents if they would feel pleased, displeased, or indifferent if their child were to marry across party lines. They found about 5% of partisans would be displeased and a similar number reported they would be pleased. Hence, 90% of partisans were indifferent (p. 97) about their child marrying someone of the out- party. One problem with this question as a measure of affective polarization is it only asks about the child marrying someone from the out-party. In order to measure affective polarization properly, one must identify those who both dislike the out-party and like their in-party. When researchers only ask about dislike for the other party, they run the risk of overestimating affective polarization. For example, Klar and Krupnikov (2016) find that respondents dislike working with someone who talks about politics even when that hypothetical colleague agrees with their political views. Further, an often overlooked confound arises in surveys when respondents infer omitted information (Dafoe, Zhang, and Caughey 2016). Some respondents will assume partisanship is an important identity for the hypothetical child-in-law if the question only mentions partisanship in describing this individual. This can increase the probability that the respondent is unhappy because the majority of Americans dislike strong partisans (Klar and Krupnikov 2016). 5

7 Finally, measuring affective polarization in the context of a larger political survey can prime partisan considerations. In particular, questions about partisan politics may bring to mind political polarization, leading respondents to believe that the hypothetical inlaw is more extreme, as Americans tend to overestimate the ideological extremity of the other party s members (Levendusky and Malhotra 2016). Research Design Data. Respondents to our two survey experiments were members of the nationallyrepresentative GfK sample via Time-Sharing Experiments for Social Sciences. The first survey took place in January The second began at the end of July, 2016 and finished on August 10, 2016 (after the Republican and Democratic Nominating Conventions). The first sample consisted of 2,030 adult Americans; the second sample of 2,136 adult Americans, 1,428 of whom had participated in the first survey. 2 The second sample allows us to evaluate how stable affective polarization is during a campaign. On the first survey, respondents also participated in an unrelated experiment about wages and the stock market. No partisan political actors were mentioned. In the second 2 GfK recruits panel members using address-based sampling. Panel members are ed when they have been assigned to a study, which may be completed online. The first survey s completion rate was 62%. For the second survey, GfK contacted 1,921 of the respondents who took the original survey (74.3% completion rate). GfK also contacted 1,208 new respondents (58.6% completion rate). Following Iyengar et al. (2012), pure independents are excluded from analyses. 6

8 survey, respondents answered no questions prior to our experiment. Hence, nothing should prime partisan considerations. Experiment. We randomly assigned both of our samples to one of three groups. 3 First, following Iyengar et al. (2012), we asked one third of the sample both of the following questions: How would you feel if you had a son or daughter who married someone who votes for the Democratic Party? Would you feel unhappy or happy? and How would you feel if you had a son or daughter who married someone who votes for the Republican Party? Would you feel unhappy or happy? 4 We provided respondents with a 5-point scale ranging from very unhappy to very happy. The second group received the same questions but with one important change: the hypothetical child-in-law was described as someone who talks about politics rarely. The final third of the sample read about a child-in-law who discusses politics frequently. These treatments eliminate the need for the respondent to infer the importance of partisanship to the child-in-law. Within each of the three experimental groups, we randomly assigned respondents to read about the hypothetical child-in-law s partisan affiliation in one of two different ways. Half read about an in-law who either supports the Democratic Party [Republican Party] and the other half read about an in-law who supports local Democratic 3 Respondents who participated in both surveys were assigned to the same treatment allowing us to look at changes over time. 4 The question order was randomized. 7

9 [Republican] candidates. This allows us to distinguish dislike for national-level parties from a dislike for any candidate affiliated with that party. Other Variables. In addition to the experimental measures, we have variables that GfK measured previously: partisanship, education levels, gender, race, and census region. These measures allow us to identify the types of individuals who are more prone toward affectively polarization. Results Happiness. Figure 1 illustrates treatment effects on level of polarization. The top graph is Wave 1 (January 2016). The bottom graph is Wave 2 (July/August 2016). [INSERT FIGURE 1 ABOUT HERE] The first panel in both graphs demonstrates the measure reported in previous studies: unhappiness with a child-in-law who supports the other party. Looking first at the control group in both waves, slightly over 30% of respondents reported unhappiness at the notion of their child marrying someone from the out- party. This is slightly greater than the percentage Iyengar et al. (2012) report for 2008, but about 10 percentage points less than their 2010 results. When the hypothetical in-law rarely discusses politics, the percentage of reported unhappiness drops by about five percentage points a decrease that is statistically discernable at the.05 level in wave 1 (one-tailed test). When the hypothetical in-law discusses politics frequently, however, there is a larger and statistically significant difference in reported unhappiness (about ten percentage points). This suggests many 8

10 individuals are more averse to disagreeable political discussion in their family than they are toward members of the other party generally. Important patterns emerge when we split the analysis based on partisan strength. Not surprisingly, weak/leaning respondents in the control group are about 30 percentage points less unhappy about partisan inter-marriage than are strong partisans. Differences in treatment effects become apparent as well: while weak/leaning partisans are only affected by the Frequently treatment, strong partisans are more affected by the Rarely treatment (the decrease is only statistically significant in Wave 1). The second panel in both graphs illustrates a previously underexplored aspect of affective polarization: happiness with their child marrying someone from their own party. The results show the vast majority of people do not care if their child marries someone from their own party: only about 35% of respondents in the control group would be happy if this occurred. That number drops below 30% in both treatment groups in both surveys (p<.05 two-tailed tests). The treatment effects are largest among strong partisans. Compared with the control group, both the Rarely and Frequently conditions lower strong partisans reported happiness with their child marrying a co-partisan. Strong partisans are less happy if their child marries someone from their party who rarely talks about politics presumably because in that situation partisanship is irrelevant. But they are also less happy if their child marries someone from their party who frequently talks about politics. Even strong partisans dislike too much political discussion -- even agreeable discussion. 9

11 Polarization. The dependent variable in Figure 2 is our best measure of affective polarization. It is a dummy variable coded 1 if a respondent is happy about their child marrying a co-partisan and unhappy about their child marrying an opposing partisan and 0 in all other cases. 5 In the control group, about 25% of respondents are affectively polarized that is, they are happier when their child marries someone from the in-party than the outparty. The control group, however, cannot distinguish people emotionally invested in partisanship from those who only want to engage in agreeable political discussion (see Huckfeldt and Mendez 2008). [INSERT FIGURE 2 ABOUT HERE] Hence, it is important to look at the Rarely treatment. If a partisan respondent gives polarized responses even when they know they will rarely have to engage in political discussion with the opposing partisan, then that person is affectively polarized. We see only about 15% of all respondents are affectively polarized in both surveys. That number is less than 10% among weak/leaning respondents and about 25% of strong partisans. In the Appendix, we estimate an ordered logit model in which we measure how polarized the individual is. We subtract happiness with their child marrying someone from the out-party from happiness with their child marrying someone from the in-party. We include additional variables to allow us to identify which types of people are polarized. 5 We replicate Figure 2 s analysis by age cohort in Appendix A4. We do not observe any systematic differences, with one exception: subjects in the age cohort are less polarized in the January survey, but not the August survey. 10

12 There are a few consistent findings. First, strength of partisanship increases polarization. Second, Republicans are more polarized, but only if they are not educated. Third, college graduates are more polarized but only if they are Democrats. The most important result confirms that previous polarization results are driven by a fear of disagreeable conversation and not pure affect, as the Rarely treatment always reduces polarization. In Wave 2 only, the Local Candidates treatment lowers the level of affective polarization, suggesting that some of the polarization in August is a reaction to a dislike of supporters of Donald Trump and Hillary Clinton and not necessarily the parties themselves. Depending on how one conceptualizes affective polarization, individuals who are responding only to presidential candidates might not be polarized because they are responding to specific political figures and not political groups. Stability of Polarization. Our final analysis looks at the stability of polarization using the measure we constructed for Figure 2 s analysis. Across all treatments, about 65% of respondents had the same level of polarization in both surveys. Of those who gave different responses, 57% became more polarized in the summer and 43% became less polarized. Table 1 s multinomial logit provides a closer look at these differences. The dependent variable has three categories: (1) the respondent is less polarized in the summer; (2) more polarized in the summer; (3) no change in polarization (the reference category). The main independent variables are the treatment variables with partisan strength interacted with both Frequently and Rarely. We control for polarization level in Wave 1. We also include a series of control variables measured by GfK prior to our survey. 11

13 Inclusion of control variables is necessary to avoid omitted variable bias with the partisan strength variable (Kam and Trussler ND), but no substantive conclusions change if we omit them (see Appendix A3). [INSERT TABLE 1 ABOUT HERE] Figure 3 presents the treatment probabilities by strength of partisanship. In the left panel, among the weak/leaning partisans, people in Control and Frequently were equally likely to be less polarized as more polarized, indicating no aggregate change. In the Rarely treatment, however, respondents were twice as likely to become less polarized as more polarized. Strong partisans in the right panel were more likely to be polarized in all three treatments. [INSERT FIGURE 3 ABOUT HERE] Interestingly, subjects who were not in the Local Candidates treatment were more likely to become more polarized. Again this may suggest measures of affective polarization often capture dislike for the partisan politics respondents see in the news instead of dislike for citizens who are Democrats or Republicans. This could also explain why weak/leaning partisans became less partisan in the Rarely treatment. Spending time with anyone even someone from the other party who will not talk about politics is a preferred. Conclusion In this paper, we argue that the extent to which modern Americans are affectively polarized may be overstated. Rather, there are two distinct phenomena that are easily conflated: affective polarization and a desire to avoid partisan politics. 12

14 Many strong partisans are affectively polarized. This can make it appear that everyone is polarized because ideologically extreme partisans are most politically engaged (Klar 2013). In the presidential campaign, they became even more polarized. But for many Americans the weak/leaning partisans - the thought of having to discuss politics with even someone from their own party is unappetizing. It is important to note these results do not imply affective polarization has not increased indeed, it has increased. It also does not mean many people actually like the other party. Rather, scholars are underestimating how much people dislike their own party. In the appendix, we analyze ANES feeling thermometers and find an increasing dislike of the in-party in recent years. These data are consistent with a theory of negative partisanship (Abramowitz and Webster 2016) individuals support their own party mainly because they dislike the other party but not consistent with affective polarization. The results in this study help improve our understanding of how affective polarization exists alongside weakening partisan identities (Klar and Krupnikov 2016). The implications of these results extend beyond how to measure polarization. Respondents in our surveys appear willing to spend time with individuals with whom they disagree as long as they do not talk about politics. The frequency of disagreement is one of the key variables in the social networks literature (e.g., Huckfeldt, Johnson, and Sprague 2004; Mutz 2006; Mutz and Mondak 2006; Ahn, Huckfeldt, and Ryan 2014). This study further demonstrates the difficulties with the conceptualization and measurement of social network disagreement (Klofstad, Sokhey, and McClurg 2013). 13

15 References Abramowitz, Alan I. and Steven Webster The Rise of Negative Partisanship and the Nationalization of U.S. Elections in the 21 st Century. Electoral Studies, 41: Ahn, T.K., Robert Huckfeldt, and John Barry Ryan Experts, Activists, and Democratic Politics: Are Electorates Self-Educating? New York: Cambridge University Press. Almond, Gabriel A., and Sidney Verba The Civic Culture: Political Attitudes and Democracy in Five Nations. Sage Publications. Bogardus, Emory S Social Distance in the City. Proceedings and Publications of the American Sociological Society, 20: Dafoe, Allan, Baobao Zhang, and Devin Caughey Confounding in Survey Experiments: Diagnostics and Solutions. Working paper available at: Groenendyk, Eric Competing Motives in the Partisan Mind: How Loyalty and Responsiveness Shape Party Identification and Democracy. New York: Oxford University Press. Huckfeldt, Robert, Paul E. Johnson, and John Sprague Political Disagreement: The Survival of Diverse Opinions within Communication Networks. New York: Cambridge University Press. Huckfeldt, Robert and Jeanette Morehouse Mendez Moths, Flames, and Political Engagement: Managing Disagreement within Communication Networks. The Journal of Politics, 70:

16 Huddy, Leonie, Lilliana Mason, and Lene Aarøe Expressive Partisanship: Campaign Involvement, Political Emotion, and Partisan Identity. American Political Science Review, 109:1-17. Iyengar, Shanto, Gaurav Sood and Yphtach Lelkes Affect, Not Ideology: A Social Identity Perspective on Polarization. Public Opinion Quarterly, 76: Iyengar, Shanto and Sean J. Westwood Fear and Loathing across Party Lines: New Evidence on Group Polarization. American Journal of Political Science, 59: Kam, Cindy D. and Marc J. Trussler. Forthcoming. At the Nexus of Experimental and Observational Research: Theory, Specification, and Analysis of Experiments with Heterogeneous Treatment Effects. Political Behavior. DOI: /s z Klar, Samara Identity and Engagement among Political Independents in America. Political Psychology, 35: Klar, Samara and Yanna Krupnikov Independent Politics: How American Disdain for Parties Leads to Political Inaction. New York: Cambridge University Press. Klofstad, Casey A., Anand Edward Sokhey, and Scott D. McClurg Disagreeing about Disagreement: How Conflict in Social Networks Affects Political Behavior. American Journal of Political Science, 57: Levendusky, Matthew S. and Neil Malhotra (Mis)Perceptions of Partisan Polarization in the American Public. Public Opinion Quarterly, 80: Mason, Lilliana I Disrespectfully Agree : The Differential Effects of Partisan Sorting on Social and Issue Polarization. American Journal of Political Science, 59:

17 Mutz, Diana C Hearing the Other Side: Deliberative versus Participatory Democracy. New York: Cambridge University Press. Mutz, Diana C. and Jeffery J. Mondak The Workplace as a Context for Cross-Cutting Political Discourse. The Journal of Politics, 68: Smith, Samantha percent of Americans now view both GOP and Democratic Party unfavorably. Pew Factank: News In The Numbers

18 Numbered List of Figure Captions Figure 1. Measuring respondent unhappiness with their child marrying someone from the other party and happiness with child marrying someone from their own party. All estimates adjusted using probability weights. A. January Survey B. August Survey 17

19 Figure 2. Subjects who are affectively polarized by partisan strength and treatment. All estimates adjusted using probability weights. A. January Survey B. August Survey 18

20 Table 1. Predicting whether respondents changed their polarization levels between January and the summer of Less Polarized More Polarized b (SE) b (SE) Rarely 0.46 (0.34) (0.27)# Frequently 0.23 (0.39) (0.25) Strong Partisan (0.46)* 0.43 (0.32) Rarely*Strong Partisan 0.52 (0.57) 1.07 (0.41)** Frequently*Strong Partisan 1.01 (0.59)# 0.35 (0.40) Local Treatment 0.67 (0.22)** (0.16) January Polarization 1.24 (0.09)** 0.11 (0.09) Republican (0.31) 0.31 (0.21) College Degree (0.32)# 0.66 (0.22)** Republican*College 0.74 (0.46) (0.33)** Male 0.49 (0.23)* 0.06 (0.16) White (0.44) 0.38 (0.41) Black (0.65) 0.06 (0.49) Hispanic (0.60) 0.16 (0.48) Midwest (0.37) (0.24)* South (0.32) (0.21) West 0.27 (0.35) 0.03 (0.24) Constant (0.65)** (0.44)** Number of Respondents 1336 A.I.C NOTE. Estimates from a multinomial logit model. Dependent variable has three categories: (-1) respondent is less polarized in the summer than January; (0) respondent has no change in polarization levels; (1) respondent is more polarized in the summer than January. The no change category is excluded as the reference. All estimates adjusted using probability weights. #p<.10, *p<.05, **p<.01 in two-tailed tests. 19

21 Figure 3. Did subjects become more or less polarized by partisan strength and treatment? Predicted probabilities calculated using values in Table 1. 20

22 Online Appendix for: Affective Polarization or Partisan Disdain? Untangling a Dislike for the Opposing Party from a Dislike of Partisanship Contents: A1. Question wording and measures for treatments and variables. A2. Ordered logit model of affective polarization levels. A3. Multinomial logit model of polarization change without control variables. A4. Replication of Figure 2 s Results by Age Cohort of Respondents A5. Using ANES Feeling Thermometers to Measure Polarization 21

23 A1. Question wording and measures for treatments and variables. Treatments (Order of Questions Randomly Assigned) Control/Not Local Candidates How would you feel if you had a son or daughter who married someone who votes for the Democratic Party? Would you feel unhappy or happy? How would you feel if you had a son or daughter who married someone who votes for the Republican Party? Would you feel unhappy or happy? Control/Local Candidates How would you feel if you had a son or daughter who married someone who voted for local Democratic candidates? Would you feel unhappy or happy? How would you feel if you had a son or daughter who married someone who voted for local Republican candidates? Would you feel unhappy or happy? Rarely/Not Local Candidates How would you feel if you had a son or daughter who married someone who votes for the Democratic Party but who RARELY talks about politics? Would you feel unhappy or happy? How would you feel if you had a son or daughter who married someone who votes for the Republican Party but who RARELY talks about politics? Would you feel unhappy or happy? Rarely/Local Candidates How would you feel if you had a son or daughter who married someone who voted for local Democratic candidates but who RARELY talks about politics? Would you feel unhappy or happy? How would you feel if you had a son or daughter who married someone who voted for local Republican candidates but who RARELY talks about politics? Would you feel unhappy or happy? Frequently/Not Local Candidates How would you feel if you had a son or daughter who married someone who votes for the Democratic Party and who FREQUENTLY talks about politics? Would you feel unhappy or happy? How would you feel if you had a son or daughter who married someone who votes for the Republican Party and who FREQUENTLY talks about politics? Would you feel unhappy or happy? Frequently/ Local Candidates How would you feel if you had a son or daughter who married someone who voted for local Democratic candidates and who FREQUENTLY talks about politics? Would you feel unhappy or happy? How would you feel if you had a son or daughter who married someone who voted for local Republican candidates and who FREQUENTLY talks about politics? Would you feel unhappy or happy? Response Options <1> Very unhappy; <2> Somewhat unhappy; <3> Neither unhappy nor happy; <4> Somewhat happy; <5> Very happy 22

24 Other Variables Partisanship Generally speaking, do you think of yourself as a Republican, Democrat, Independent, Another Party, or No preference? 6 (If answered Republican) Would you call yourself a strong Republican or not a very strong Republican? (If answered Democrat) Would you call yourself a strong Democrat or not a very strong Democrat? (If answered Independent, Another Party or No preference) Do you think of yourself as closer to the Republican Party or Democratic Party? Republican is coded 1 if respondent answered Republican or closer to the Republican Party and 0 if the respondent answered Democratic or closer to the Democratic Party. Pure independents are eliminated from the analysis as they cannot be assigned to their own party or the other party in the treatment questions. Strong Partisan is coded 1 if the respondent said they were a strong Republican or a strong Democrat and 0 if the respondents said they were not a very strong partisan or that they were an independent in response to the first question. Education College coded 1 if highest degree completed is Bachelor s Degree or higher and 0 if highest degree is high school, some college, or less than high school. Gender Male is coded 1 if the respondent s self-reported gender is male and 0 if the respondent s selfreported gender is female. Race White is coded 1 if the respondent s self-reported race is White Non-Hispanic and 0 if the respondent s self-reported is Black Non-Hispanic, Other Non-Hispanic, Hispanic, or 2 or more races Non-Hispanic. Black is coded 1 if the respondent s self-reported race is Black, Non-Hispanic and 0 if the respondent s self-reported is White Non-Hispanic, Other Non-Hispanic, Hispanic, or 2 or more races Non-Hispanic. 6 This question was provided by GfK and is used as part of the GfK panel measures. The wording of this question is slightly different than that of the ANES partisanship measure; while the ANES measure begins with Generally speaking, do you usually, the GfK measure excludes the word usually (as shown above). The question still however, retains an important commonality with the ANES question by starting with the words generally speaking. As Abramson and Ostrom (1991) write, a partisanship question that begins with generally speaking rather than one that begins with in politics as of today or in politics today is more likely to get at longer term partisan predispositions. Moreover, MacKuen et al. (1992) use a New York Times and CBS poll question that also excludes the word usually, but retains generally speaking, and argue that it is equivalent to the ANES question. 23

25 Hispanic is coded 1 if the respondent s self-reported race is Hispanic and 0 if the respondent s self-reported is White Non-Hispanic, Black Non-Hispanic, Other Non-Hispanic, or 2 or more races Non-Hispanic. Census Respondents are coded as Midwest if their state of residence is Illinois, Indiana, Iowa, Kansas, Michigan, Minnesota, Missouri, Nebraska, North Dakota, Ohio, South Dakota, or Wisconsin. Respondents are coded as South if their state of residence is Alabama, Arkansas, Delaware, Florida, Georgia, Kentucky, Louisiana, Maryland, Mississippi, North Carolina, Oklahoma, South Carolina, Tennessee, Texas, Virginia, or West Virginia. Respondents are coded as West if their state of residence is Alaska, Arizona, California, Colorado, Hawaii, Idaho, Montana, Nevada, New Mexico, Oregon, Utah, Washington, Or Wyoming Respondents are coded as Northeast (reference category) if their state of residence is Connecticut, Maine, Massachusetts, New Hampshire, New Jersey, New York, Pennsylvania, Rhode Island, or Vermont. 24

26 A2. Ordered logit model of affective polarization levels. Dependent variable is coded: (happiness with own party)-(happiness with other party) If it is coded 0, the respondent is happier with their child marrying someone from the other party or equally happy with either outcome. Values greater than zero indicate greater happiness with child marrying own party and therefore greater polarization. January Survey August Survey Control Rarely Frequently Control Rarely Frequently % 70.9% 61.9% 58.0% 71.7% 67.4% 1 7.8% 9.2% 12.0% 8.6% 7.0% 9.2% % 9.8% 10.9% 12.3% 10.3% 10.9% % 4.5% 7.66% 10.8% 5.0% 6.4% 4 8.9% 5.5% 7.5% 10.4% 6.0% 6.1% N January Survey August Survey Coef. Std. Err. Coef. Std. Err. Rarely ** *** Frequently Strong Partisan *** *** Rarely*Strong Partisan Frequently*Strong Partisan ** Local Candidates ** Republican * *** College Degree * *** Republican * College ** Male White Black Hispanic Midwest South ** West τ τ τ τ N AIC

27 Predicted values from model January Survey August Survey 26

28 A3. Multinomial logit model of polarization change without control variables. Less Polarization More Polarization Coef. Std. Err. Coef. Std. Err. Rarely Frequently Strong Partisan ** * Rarely*Strong Partisan Frequently*Strong Partisan * Local Treatment *** ** January Polarization *** Constant *** *** N 1336 AIC This is a replication of Table 1. All of the estimated probabilities in Figure 3 change by less than 1 percentage point based on this model indicating the inclusion of the control variables does not affect our estimates of treatment effects by strength of partisanship. 27

29 A4. Replication of Figure 2 s results for all respondents by respondent s age cohort. 28

30 A5. Using ANES Feeling Thermometers to Measure Affective Polarization In examining, the ANES feeling thermometer ratings we look only post-1978 and only at the ratings of the Democratic Party and Republican Party. We do this because prior to 1978 the ANES asked respondents to rate Democrats and Republicans and not the parties. We do not want to attribute changes to different eras if those changes are really the result of different question wordings. Also, we examine the ratings of the parties and not particular presidential candidates because the ratings would also not be comparable across time given that the changes might be attributable to specific, non-partisan factors of the individual candidates. The next figure at the top of the next page looks at the average in-party (left panel) and out-party (right panel) feeling thermometer ratings by year. The solid lines represent strong partisans while the dashed lines represent weak/leaning partisans. Among the strong partisans the in-party feeling thermometer ratings have largely remained unchanged--though, 2016 does have the lowest rating recorded. More dramatic is the change in the out-party from above 40 in 1980 to less than 20 in the last two cycles. This is strong preliminary evidence of polarization among strong partisans. 29

31 What about the weak/leaning partisans? While feeling thermometer ratings of the out-party among weak/leaning partisans have declined, the feeling thermometer ratings for the in-party have also declined in the last two cycles. In 1988, the in-party feeling thermometer peaked at 71. In 2016, the rating was 58. This decline is not as sharp as the decline in the out-party ratings, but it is still meaningful. According to the labels the ANES gives to these ratings on the thermometer, the average weak/leaning partisan in 1988 had fairly warm feelings towards his or her party. By 2016, the typical weak/leaning partisan s rating of his or her own party was less than a bit more warm than cold. A more direct measure of polarization is the difference between in- and out-party feeling thermometer ratings. We present this measure in our next figure. Again, we separate between strong and weak/leaning partisans. 30

32 Importantly, both groups show increased polarization in recent years. Among strong partisans, the average difference was around 40 points at the beginning of the time series. Currently, it is around 60 points. Among weak/leaning partisans, the average difference was about 20 in the 1980s while it is currently around 30 points. Hence, while both groups have become more polarized on this measure, the increase in polarization is twice as large among strong partisans. While differences in feeling thermometer scores are a better measure of polarization than measures that only look at the out-party ratings, they are not the best measure possible of polarization. For example, consider a difference of 30 points the average distance for weak/leaning partisans. That could result from rating the out-party 40 and the in-party 70. One could state that respondent was polarized he or she is a bit more cold than warm towards the other party and fairly warm towards their own party. Certainly, this is not the most polarized an individual could be, but at least they like their party and dislike the other party. On the other hand, a difference of 30 could also come from someone who rates the out-party 0 and the in-party 30. This individual is not polarized he or she dislikes one party more, but he or she does not like either party. 31

33 To address polarization even more directly, we calculate a new measure in our final figure. With this measure, a respondent is polarized if the respondent both gives the inparty a high rating and the out-party a low rating. We set two different thresholds. The panel on the left has a less stringent threshold (60 or above for the in-party; 40 or below for the out-party) while the panel on the right has a more stringent threshold (70 or above for the in-party; 30 or below for the out-party). Beginning with the less stringent threshold, we see that the percentage of weak/leaning partisans who have become polarized has increased. It was around 30% for most of the 1980s and was around 40% in the early 21 st century and 46% in This increase, however, is much smaller than the increase among the strong partisans. The percentage of polarized strong partisans was less than 50% in both 1978 and It was above 80% in 2012 and This is probably categorizing too many respondents as polarized. To be polarized at this threshold, one only needs to be a bit more warm to their own party and a bit more 32

34 cold to the out party. That is why we set the greater threshold in the right panel. Again, polarization among both groups increases over time, but the increase is much greater among the strong partisans. The percentage of polarized strong partisans increases from about 30% to 60%. The percentage of polarized weak/leaning partisans increased from about 10% to 20%. Hence, if we look at the ANES feeling thermometer ratings we see a 40 percentage point difference in the level of polarization between weak and strong partisans. This is almost exactly the same difference we see in our data. Hence, whether we look at the marriage question or ANES feeling thermometer ratings, we see the exact same difference in polarization between strong and weak/leaning partisans. 33