Katerina Linos, J.D./Ph.D.,

Transcription

1 Citizen Responses to the Supreme Court s Health Care and Immigration Rulings: Comparing Experimental and Observational Methods Can the U.S. Supreme Court change Americans policy views? We surveyed a representative sample of Americans before and after two major Supreme Court decisions, embedding an experiment in the second survey. In contrast to much of the prior literature, we find that the Supreme Court can lead public opinion, even on hotly contested issues such as health care and immigration. Respondents who were offered clear messages in support of the majority s reasoning reacted much more positively than respondents also reminded about the dissent. In addition, trust in the Court, partisan affiliation, and ethnic background mediated responsiveness. These findings invite us to revisit debates about the role of an unelected Court in a democracy. They also contribute to debates on survey methodology, by highlighting that both experimental and real- life treatments can yield similarly sized effects for the population at large when events receive broad news- coverage and experimental treatments resemble news headlines. Katerina Linos, J.D./Ph.D., Assistant Professor UC Berkeley Law School klinos@law.berkeley.edu Kim Twist Ph.D. Candidate Dept. of Political Science UC Berkeley kim.twist@berkeley.edu Draft Comments very welcome! We thank Kathy Abrams, Jack Citrin, Maria Echaveste, Chris Edley, Stavros Gadinis, Stephen Goggin, Tom Goldstein, Taeku Lee, Gabriel Lenz, Justin McCrary, Victoria Plaut, Alison Post, Kevin Quinn, Bertrall Ross, David Sklansky, Laura Stoker, Cecillia Wang, and Chuck Weisselberg for their generous advice. We are also very grateful to the Hellman Family Fund and the Warren Institute for financial support, and to Annie Hilby and May Whitaker for excellent research assistance. We also thank Samantha Luks and Ashley Grosse at YouGov/Polymetrix for their help in fielding the surveys.

2 Introduction Can the Supreme Court, a highly esteemed institution, change citizens views so they support its decisions? Or do the Court s choices mainly clarify and reinforce citizens prior beliefs? As the Supreme Court takes on new and controversial questions, including health care reform and gay marriage, the time has come to revisit this powerful institution s influence. Although the Court may fall in line with public opinion (Epstein & Martin 2010; Casillas et al. 2011), when opinion is deeply divided, the Court s ability to move that opinion in one direction or the other is of great importance. Earlier studies on the impact of the Court reach conflicting conclusions. Observational studies, which compare public opinion before and after actual Court decisions, often find that the Court has no net impact on Americans views (Marshall 1987; Rosenberg 1991; Hanley 2008; Le & Citrin 2008; Luks & Salmone 2008; Gash & Gonzalez 2008), although the Court can polarize opinion and push different groups in opposite directions (Franklin & Kosaki 1989; Johnson & Martin 1998). Experimental studies, in which some respondents are informed about a (usually fictitious) Court decision from the experimenter, often find that the Court can lead public opinion (Bartels & Mutz 2009; Hoekstra 1995; Clawson et al. 2003). This difference could occur because actual Supreme Court decisions often come down after controversies have received significant media exposure, after Americans have thought about the issues and developed firm views. Alternatively, these conflicting findings could result from severe methodological limitations. Earlier observational studies are based on survey questions fielded for other purposes. On issues such as flag burning, private property takings, and even desegregation, we know what Americans thought following critical Court decisions, but we have no baseline against which to compare these views (Rosenberg 1991; Murakami 2008; Hanson 2008; Nadler et al. 2008). Even when before and after data exists, the after data was often collected long after the Court decision; events other than the Court decision could influence opinion in this interval. For example, in Thomas Marshall s pioneering research, considered the most comprehensive observational study to date, the average time lag between the Court decision and the post- decision survey exceeded two years (Marshall 1987: 165; Egan & Citrin 2011: 6). Moreover, in some observational studies, the wording of the after survey did not match either the wording of the before survey or the exact issue before the Court (e.g., Johnson & Martin 1998: 307). External validity limitations that often plague experimental studies could also account for the difference between observational and experimental studies. For example, some experiments on the influence of the Court survey college students, who tend to be younger, better educated, and more liberal than Americans in general (e.g. Unger 2008: 758; Hoekstra 1995). Because each of these factors is linked to greater- than- average confidence in the Court (Stoutenborough and Haider- Markel 2008: 38), Court decisions could influence students more than the general population. More critically, Jason Barabas and Jennifer Jerit (2010) recently cast significant doubts on the entire field of experimental survey research by 1

3 reporting that real world events had no effects on the population at large, whereas similar experimental treatments yielded large shifts. Barabas and Jerit attributed this difference to the fact that in survey experiments, everyone in the treatment group is exposed to information, whereas in the real world, few people follow the news (2010: 226). By studying events that received much greater coverage than those examined by Barabas and Jerit, we further develop this research. We investigate whether exposure is the critical difference between experimental and observational studies, or whether other factors are at play, such as an experimenter s ability to make particular considerations salient to respondents (Zaller 1992: 49). Our research design combines observational and experimental techniques and builds on the strengths of prior studies while also addressing these important limitations. We surveyed a representative sample of Americans shortly before and shortly after two major Supreme Court decisions. As the Court decisions focused on health care and immigration, we asked respondents for their views on these two issues and their confidence in the Court. Because the health care decision upheld a law championed by Democrats, while the immigration decision upheld a law championed by Republicans, we were able to examine how people responded both to a liberal and to a conservative decision. We interviewed the same people at two points in time, which greatly enhances the precision of our estimates (Mutz 2011). In addition, if some people respond by moving towards the Court s position, while others respond by moving away from it, as prominent theories of backlash and polarization suggest (Franklin & Kosaki 1989; Johnson & Martin 1998), re- interviewing the same people allows us to capture all of this movement. To complement the before and after comparison, we embedded an experiment in the second, post- decision survey to examine whether the type of information individuals receive about Court decisions influences their responses. Some Supreme Court decisions are unanimous and offer citizens a single frame through which to understand an issue. Others contain powerful dissents, thus offering a second, competing frame. Political communication studies suggest that while one- sided messages can have large effects on public opinion, competing messages can neutralize one another (Chong & Druckman 2007, 2013). To examine whether dissenting arguments dilute the influence of the Court s majority opinion, we randomly assigned some respondents to be reminded of the main argument the majority deployed to support its holding, while others were also offered the main argument of the dissent. In addition, we monitored what information participants received about the Court decisions through the media sources they normally use. We found that Supreme Court rulings can change people s attitudes and lead public opinion, even on highly politicized and extensively debated issues such as health care and immigration. Moreover, both experimental and real life treatments can yield similarly sized effects when events receive broad news coverage and experimental treatments resemble news headlines. In the case of the health care decision, which received very broad media coverage, large portions of the population correctly understood the holding and moved towards the court s position even without a reminder from us. Moreover, by repeating the holding of the decision, and the main argument of the majority in the course of the experiment, we 2

4 did not further increase the size of this shift. This suggests a true information effect at work, and a possible limit to experimental manipulations. We also found that respondents exposed to one- sided messages in support of a decision were significantly more likely to change their views in the direction of the Court than respondents exposed to two- sided messages. Indeed, we found similarly sized effects among respondents who we randomly assigned to the health care case dissent and among respondents who were exposed to the dissent through Fox News, which gave the dissent particular prominence. We also found that the Court s influence was moderated by several factors. People with high trust in the Court before the decisions came down were more likely to change their minds and follow the Court ruling after the decisions. Independents shifted their views to follow the Court in both cases. Responses to the highly politicized health care decision followed partisan lines. Democrats became much more supportive of the individual mandate than they had been before the Court decision, while Republicans mostly maintained their earlier opposition. We also noted backlash in the context of the immigration decision. Hispanics, especially those exposed to critical coverage of the decision through Spanish- language television, became much more critical of the key immigration restriction that the Supreme Court upheld. While our study is uniquely able to investigate all this variation in responsiveness to the Court, we are especially confident in our results because independent polling by several major organizations confirms that overall support for the Affordable Care Act increased after the Supreme Court decision, and that this increased support has persisted in subsequent months (Figure 4 below; Campbell & Persily 2012). In addition to their methodological implications, these findings also have implications for important debates about the role of the judiciary in our democracy. The Supreme Court enjoys unusually high levels of trust among the American public (Gibson et al. 2003). In addition, Supreme Court justices control which cases they review, when they issue decisions, what holding they reach, whether they issue unanimous decisions or not, and how broadly or narrowly they write their opinions. Some scholars suggest that the Court should use its broad discretion to avoid engaging with politically controversial issues (Bickel 1986; Schauer 2006: 9). Others argue instead that Courts should try lead public opinion to avoid a tyranny of the majority (Ely 1980; Dorf 2010). The health care decision brought this controversy in particularly sharp relief, as the Court chose to rule on the Obama administration s signature legislative achievement only months before a national presidential election. Empirical data on whether the Court influences public opinion, whose opinion it influences, and how dissents shape this effect are critical to developing these major normative debates (Persily et al. 2008: 5). The Court s Influence on Opinion in Experimental and Real- Life Settings We argue that we can best understand and measure the influence of a Court decision by separating the process into three analytical components: (1) whether a person hears about a decision and understands its main holding; (2) what she hears about the decision, and more specifically, whether she hears one- sided or two- sided 3

5 arguments; and (3) what individual characteristics, such as partisanship and trust in the Court, may mediate her responsiveness. Breaking the process into these components also allows us to contribute to the literature on the external validity of experimental treatments. Barabas and Jerit (2010) conclude that real- world government announcements, unlike comparable experimental announcements, often have no effect on the population at large, because news coverage is limited. We test whether real life- announcements that are extensively covered in the media yield large population wide- effects, and whether experimenters can still manipulate these effects by making particular considerations salient to respondents. First, for an event to influence a person s opinion, she must hear about it (e.g. Huber & Arceneaux 2007). Figure 1, below, shows that Supreme Court decisions can generate very high levels of news- coverage. Following Barabas and Jerit (2010), we searched the LexisNexis archive of major US newspapers for "health care" and "immigration," and for both terms mentioned with either presidential candidate's name, for the six months preceding the 2012 elections. Both issues were salient throughout the campaign; our search averaged 88 and 29 daily stories for health care and immigration, respectively. 1 In contrast, the Barabas and Jerit study reports just over fifty stories total for each of their two examples over a five- week period. 2 Media coverage of both health care and immigration spiked after the Supreme Court decisions to levels not seen before or since. No other actor or event, including the presidential debates, placed as much attention on these issues as the Supreme Court decisions FIGURE 1 HERE - - Table 1, below, presents the volume of news coverage generated by the two Supreme Court cases we studied, compared to other important Supreme Court cases and to the two events studied by Barabas and Jerit. The tables report the number of stories in the New York Times that referenced each case in the three weeks before and after both the oral arguments and the decisions. 3 Many Court cases received orders of magnitude more coverage than the events studied by Barabas and Jerit. Indeed, it is possible that coverage of Supreme Court cases might be increasing over 1 These totals include 29 different news sources. 2 Their examples focused on an announcement that the Medicare trust fund would run out, and a change in the test that immigrants must take to become US citizens. 3 We found that more than seventy percent of coverage around Supreme Court cases happens around the date of the decision, with the remainder occurring around oral argument. To find coverage for the events studied by Barabas and Jerit, we replicated the search using the terms citizen and Medicare, to cast the search net as widely as possible, but found only two stories for each of these events in the relevant time period. Though the New York Times tends to be more liberal than other major publications (Ho & Quinn 2008), we believe that Table 1 gives us a helpful first cut about the relative publicity different events received. This measure correlates highly with another measure of case publicity, references in the Readers Index to Periodical Literature. 4

6 time, when taking into account the topic of the decisions. For example, Table 1 suggests that early abortion cases (Roe) received less coverage than later abortion cases (Webster, Casey); early affirmative action cases (Bakke) received less coverage than later affirmative action cases (Grutter and Gratz); early campaign finance cases (Buckley) received less coverage than later campaign finance cases (Citizens United); and early gay rights cases (Bowers) received less coverage than later gay rights cases (Lawrence). Despite these variations, because Supreme Court decisions often receive extensive coverage, we expect that many Americans should be able to correctly identify the Court s position. While an older literature suggests that Americans are generally unaware of news developments (e.g. Delli Carpini & Keeter 1996), more recent work (e.g. Gibson & Caldeira 2009b) suggests that people are more aware than we had believed, and more aware of the Court, in particular. We follow the advice of Gibson and Caldeira (2009b) in asking closed- ended questions to measure knowledge, and expect to find levels of awareness comparable to their findings. In addition, because the Court enjoys citizens trust (Caldeira & Gibson 1992; Gibson & Caldeira 2009a; Hoekstra & Segal 1996), we expect the following hypothesis to hold true: Hypothesis 1: Supreme Court decisions that receive widespread coverage should influence Americans views and move aggregate public opinion in the direction of the Court s ruling, even in the absence of experimental reminders TABLE 1 HERE Second, what people hear about a decision should matter. Sometimes, Supreme Court decisions are presented as the final resolution of the conflict before the Court; news media often present the only Court majority s view. Often, however, Court decisions have vigorous dissents, and these dissenting views also get extensive news coverage (Ho & Quinn 2008). There are no prior studies that focus on competing argumentation in the context of Court opinions. 4 However, an important prior experimental study suggests that merely mentioning that a decision was unanimous or divided, without any argumentation, can change the impact of the Court on opinion (Zink et al. 2009). Chong and Druckman (2007, 2013) report that experiments with competing frames in other political contexts yield much smaller shifts in opinion than experiments involving one- sided frames. As part of our experiment, some respondents were reminded of the argument put forth of the Court s majority (Reminder 2), while others were also reminded of the argument of 4 Simon and Scurich (2011) examine argumentation in a different context. They compare nuanced majority opinions (i.e. opinions acknowledging some counter- arguments) to majority opinions alone, finding that nuanced majority opinions in some cases increase the acceptability of Court decisions. In contrast, we expect and find that dissents decrease the persuasive influence of the Court. 5

7 the dissent (Reminder 3). 5 We are also interested in comparing this experimental manipulation to similar information some respondents likely received through real- world news- sources. Most of the media coverage of the health care decision emphasized the majority s position. A typical headline was "Supreme Court upholds health care law" (The Washington Post and The Boston Globe, June 28) or ""High Court backs health care law" (The Chicago Tribune, June 28). However, Fox News heavily emphasized the dissent, saying, The majority is rewriting the statute and making it a tax. 6 As a result of this news coverage, we might expect to see reduced responses to the Court s majority opinion not only among persons reminded about the dissent in the course of the experiment, but also among Fox News viewers. Hypothesis 2: Respondents who are exposed to a one- sided frame from the Court s majority should follow the Court more than respondents who are also exposed to a competing frame from the dissent. Third, individuals may respond in varied ways to the messages they receive. Druckman (2001), among others, reports that, in other contexts, credible sources are more persuasive. Therefore, we would expect individuals who expressed high levels of confidence in the Court before the decisions to be particularly likely to change their opinions and follow the Court decisions. We also investigate whether people who followed the Court s lead on the issues of health care and immigration were more likely to retain their trust in the Court after these decisions. In addition, we expect partisanship to mediate responses (Cobb & Kuklinski 1997; Lee & Schlesinger 2001; Lenz 2009). We expect Independents to be especially likely to follow the Court, because they are less likely to have formed firm opinions prior to the decision based on partisan cues. The health care case was highly politicized: President Obama strongly praised the Court s decision on the individual mandate, while Governor Romney, the Republican presidential nominee, sharply criticized it (Fox News Online 2012a, 2012b). In contrast, statements following the immigration decision were more limited and mixed in tone. 7 We therefore expected large shifts 5 Other respondents received no reminder, while still others were only offered the Court s holding, with no arguments (Reminder 1). 6 On the Record with Greta Van Susteren, June 28. Similarly, Sean Hannity began his Fox program that evening with the line, Today will be remembered as the day that the Supreme Court of the United States of America upheld the largest tax increase in American history ( Hannity, June 28). In contrast, MSNBC hosts began their evening shows by discussing one of the great days in this country s history, ( Hardball, June 28) and the historic news day ( The Rachel Maddow Show, June 28). The evening news programs on both CBS and NBC were measured in their recounting of the day s events; neither host offered a pro- or anti- mandate viewpoint, and both shows presented videos and quotes from Republicans and Democrats alike. 7 President Obama issued a statement in support of the Justices decision to strike down the other aspects of the law, but spoke out about other concerns of racial 6

8 among Democrats following the health care ruling, because they received positive endorsements both from the Court and from party elites, while we expect more limited movement among Republicans, who received both a positive message from the Court and a critical message from party elites (including Fox News, which focused primarily on the dissent s arguments). Hypothesis 3: Respondents who report high levels of trust in the Court before the decisions, Independents, and people receiving pro- Court messages from party elites should be particularly likely to follow the Court. Finally, earlier studies of the Court highlight the possibility of backlash: that subgroups of Americans become more hostile to Court s position following a decision. Backlash could result if a Court decision brings attention to an issue many citizens had not carefully considered previously (Johnson & Martin 1998: 299). For example, Franklin and Kosaki (1989: 762-3) report that, following Roe v. Wade, Catholics opposition to some types of abortions solidified. Such opinion shifts are especially likely among groups that the decision impacts directly (Hoekstra & Segal 1996) and among groups that receive critical coverage of the Court decision. Based on the prior literature, we would anticipate backlash among Hispanics following the Court s immigration decision, which upheld an important Arizona restriction. The Court decision placed Arizona s immigration restrictions in the national spotlight (see Figure 1 above). And in Spanish- language media, much of the coverage was negative. For example, on the day of the decision, the headline of La Opinión, the country s largest Spanish- language daily, was The Arizona Community is Full of Doubts, and both Univision and Telemundo aired thirty- minute, commercial- free segments on the decision after it was handed down. 8 We therefore expect to see some negative responses to the Court decision among Hispanics, especially among Hispanics who follow Spanish- language media. Hypothesis 4: Citizens may move away from the Court s position if they are exposed to large amounts of critical news coverage about a Court decision. Research Design We fielded a two- wave survey shortly before and shortly after two important Supreme Court decisions, embedding an experiment in the second wave. 9 Extensive news coverage and public opinion surveys often follow Court decisions, but the profiling that remain. In contrast, Mitt Romney made a more general statement blaming the Obama administration for having not handled the immigration issue earlier (Gabriel & Cooper 2012). 8 Fernández (2012). Similarly, coverage on Univision and Telemundo was very critical. See, for example, Univision s video archive, /buscar /resultados.do?search_type=video&query=sb YouGov/Polymetrix fielded the first wave in mid- May 2012, and the second wave in late June 2012, starting on the day after each decision was announced. 7

9 scarcity of survey data fielded before key decisions has limited prior research. The Supreme Court announces the cases it will review early on. However, until a decision comes down, there is significant uncertainty about every aspect of the decision. For example, we do not know whether the Court will reach the merits of a case or dispose of it on procedural grounds, when the decision will come down, or which way the Court will rule. By consulting with several lawyers who closely followed the Court and participated in related litigation, we were able to somewhat limit this uncertainty and field questions about the major substantive provisions of each challenged law before the Court decisions came down. Our opinion questions examine the individual mandate provision of the Affordable Care Act and the show your papers provision of Arizona s immigration law. 10 That said, the direction of the decisions came as a surprise to many observers. On the eve of the widely anticipated health care case, futures markets showed more than a seventy percent chance that the individual mandate would be struck down, a prediction that was ultimately incorrect (Bingham 2012). A majority of constitutional law professors surveyed before this decision also incorrectly predicted its outcome (Drummond 2012). This reassures us that, though there was significant publicity surrounding the health care case before the decision, even experts who follow Court news closely received new information on the constitutionality of the individual mandate on the day the Court decision was announced. We fielded the second wave of our survey days after each decision was announced. Long lags between question waves are an important limitation of earlier studies (Hoekstra 1995: 112). The short time frame between the waves, as well as the proximity of the decisions to the second wave, gives us more confidence that the Court s rulings, rather than other events, drive our results. That said, we are also interested in whether any short- term shifts persist, a point that we investigate below by employing poll data from other sources. We interviewed the same respondents before and after the decisions. Such within- subjects designs are often preferable to between- subjects designs, because they yield more precise estimates. However, a concern with these designs is that respondents will catch on to the manipulation. Mutz (2011: 93-4) offers nationally representative internet- based surveys, such as the one we used, as a useful solution to this problem, largely because respondents receive multiple surveys, and are thus unlikely to be strongly influenced by questions on any one such survey. Indeed, given Bartels (1999) finding that panel effects rarely occur, the short length of the Wave 1 survey, and the fact that it included only one question on the Supreme Court, we believe it is unlikely that our Wave 1 questionnaire influenced respondents views six weeks later, when they responded to the Supreme Court decisions and completed Wave 2. At Time 1, survey respondents were randomly assigned to receive either the immigration or the healthcare arm of the survey. Both sets of respondents were asked Q1 (personal importance of immigration and health care), Q2 (news sources), and Q4 (confidence in the court), while Q3 (opinion on immigration or health care) 10 See Appendix 1 for all question wording and response options. 8

10 varied in each survey arm. In drafting these questions, we followed earlier studies where possible to enhance comparability. To ask about the personal importance of immigration and health care, we followed the phrasing used by YouGov/Polymetrix in past surveys. We drew on Egan and Citrin (2011) for our question on confidence in the Court. Our media questions focused on television stations because the vast majority of Americans still get most of their news in this way and because we expect the most powerful and partisan stimuli to come through television news (Pew Research Center 2012). For the second wave of questions, fielded days after the health care and immigration decisions were released, respondents were asked about the same topic as in Wave 1 (health care or immigration). To disguise our purpose, and to avoid reminding respondents that they may have answered similar questions six weeks prior, Wave 2 began with a series of news headlines and asked respondents which, if any, they had seen. If respondents ticked the news- headline corresponding to the Supreme Court decision, they were then prompted to answer 1-3 questions about the Court decisions. Following Gibson and Caldeira s (2009b) recommendations, we asked closed- ended rather than open- ended questions, as these more accurately capture knowledge about the Court. These knowledge questions allow us to add data to a debate about how much people know about Court decisions (Gibson & Caldeira 2009b). After completing these questions, some respondents received information about the Court decision. Respondents were randomly assigned to one of four groups while blocking on race, interest in the news, and viewership of Fox news in particular. Blocking helped us to minimize error and reduce noise in our results (see Mutz 2011: 95). For each experiment, forty percent of respondents were assigned to the no- reminder group (n=400), and these people did not receive any information about the content of the Court decision from the experiment. The remaining respondents were divided into three equally sized groups and shown different reminders about the decision (n=200 per group). Our reminders drew on the major arguments developed by the Justices (see Appendix for the complete text). In the health care decision, there was both a powerful majority opinion and a forceful dissent, so we drew our arguments in favor and against the individual mandate from these opinions. Reminder 1 offered the main holding of the decision, Reminder 2 offered this holding plus an argument favor of the decision from the Court majority, while Reminder 3 also an argument against the decision from the Court s dissent. The immigration ruling, however, was unanimous on the key provision of the law. We modified our design accordingly, and emphasized the fact that this was a unanimous decision in Reminders 2 and 3 (but not in Reminder 1) of the immigration experiment. Even though there was no dissenting opinion in the immigration case, we added a consideration from the decision that pushed against the majority s conclusion in Reminder 3. Prior research on nuanced majority opinions suggests that this additional consideration could work quite differently from a dissent, and in some circumstances might increase, rather than reduce, the influence of the majority s position (Simon & Scurich 2011). Results 9

11 We want to know whether people changed their minds about the individual mandate and the Arizona immigration restrictions following the Court decisions. Thus, our dependent variable takes on three values: a 1 for an increase of support in each provision from Wave 1 to Wave 2, 0 for no change, and - 1 for a decrease in support. Because the Court upheld both laws, a 1 can also be interpreted as a shift in the direction of the Court ruling, while a - 1 can be interpreted as a shift away from the Court ruling. This coding allows for comparability with earlier studies and lessens the impact of outliers. 11 Because this measure of change is based on the attitudes of the very same individuals at two points in time, we have a great deal of confidence in our descriptive statistics, which are very consistent with our regression results TABLE 2 HERE Table 2 presents these first descriptive statistics. The first column shows whether people shifted their views, broken down by issue (health care or immigration) and by the group to which respondents were randomly assigned (no reminder, holding only, holding + argument for, holding + arguments for and against). For example, the first row of the first column indicates that following the health care ruling, a net 9.5 percent of respondents who received no reminder shifted toward the Court s position, a shift different from zero at the 0.01 level. 12 Because this group received no reminder from us, we would expect this effect to be driven primarily by people who were aware of the Court decision, likely having heard through the media. We thus broke down the sample depending on whether respondents reported a high interest in following the news before the decisions 11 Egan and Citrin (2011), an experimental study that employs panel data, also used this coding. We examined an alternative coding, which gives greater weight to respondents who changed their minds radically. Because we asked respondents to express their degree of support or opposition on a 5 point scale at two points in time, we also constructed a nine- point scale ranging from - 4 to +4, where - 4 represented a move from strong support in Wave 1 to strong opposition in Wave 2, and 4 represents the reverse. Results were generally similar across the two measures. However, we worry that few outliers who moved from one extreme to the other on our survey may have made a mistake or rushed through the survey. We thus report results that give these outliers the same weight as all other respondents. 12 More specifically, 66.5 percent of our 382 respondents in this condition did not change their views from Wave 1 to Wave 2, 12.0 percent moved away from the Court s position, and 21.5 percent moved towards the Court s position. Our coding includes people who changed their minds slightly for example, people who moved from opposing the individual mandate strongly to opposing the individual mandate somewhat. If we look only to the change in support for the mandate, we see a 6 percent increase, from 29 percent support in Wave 1 to 35 percent in wave 2, for the group that received no reminder from us. 10

12 came down. 13 In both the health care and the immigration experiments, we see a statistically significant shift towards the Court among people with a high interest in the news, i.e., people likely to be aware of the decisions, and no such shift among people less interested in the news. This observation, confirmed in subsequent regression analyses, supports Hypothesis 1, that Supreme Court decisions that receive widespread coverage influence American s views and move aggregate public opinion in the direction of the Court s ruling, even in the absence of experimental reminders. That said, in the no- reminder groups, the shift following the health care decision was much larger than the shift following the immigration decision. This is likely in part because the health care case received more news coverage than the Arizona decision. In addition, news coverage of the Arizona case was particularly confusing, because the Court upheld the main provision of the Arizona immigration law, while striking down other provisions. 14 As a result, 46 percent of our respondents correctly noted that the Court upheld the key Arizona immigration restriction we studied, 14 percent incorrectly believed it had been struck down, and another 40 percent did not know which way the Court had ruled. In comparison, 70 percent of respondents correctly identified that the Court upheld the individual mandate, 6 percent incorrectly believed that the law had been struck down, while another 25 percent did not know. 44 percent of all our respondents not only correctly understood that the Court had upheld the law but were able to answer two additional more detailed questions about the case. 15 These findings are consistent with Gibson and Caldeira s findings that Americans are quite knowledgeable about the Supreme Court (2009b). 13 We also have more detailed data on whether people correctly identified the holding of each decision or not. The two measures are correlated (r=0.4) and results are very similar using both methods. However, conservatives were heavily overrepresented among people who correctly noted that the Court upheld Arizona s immigration restrictions. Thus, we focus on general interest in the news for the cross- tabulations, and return to measure of knowledge of the decision in the regression models, where we can better control for party affiliation and other confounders. 14 Opportunity Agenda (2012) systematically reviews this confusing coverage; some examples follow. In the hours following the announcement, the New York Times website was one of just a few sites to suggest a favorable ruling, saying, "Justices uphold key part of Arizona law" (June 25). In contrast, the CNN headline announced a "blow to immigration law," with the Court telling Arizona "you went too far," and a more reserved Fox News website said that the "Supreme Court reigns in Arizona on immigration" (June 25). The vast majority of the coverage was somewhere in the middle, with headlines like "Court rejects parts of Arizona immigration law" (The Washington Post, June 25), "High Court repeals most of Ariz. immigration law" (The Baltimore Sun, June 26), and a more- detailed "Supreme Court rejects most of immigration law; The ruling is largely a victory for Obama, but the court leaves in place a 'show me your papers' provision" (The Los Angeles Times, June 26). 15 See appendix for question text. 11

13 We next investigated whether we could manipulate these effects by making particular considerations more salient to respondents. Specifically, we sought to investigate whether respondents exposed to one- sided arguments were more likely to follow the Court as compared to respondents exposed to two- sided arguments. Chong and Druckman s (2007, 2013) experimental research indicates that strong counter- frames can significantly reduce the effect of positive endorsements, but that weak counter- frames do not have this effect. Table 2 suggests, and subsequent analyses confirm, that, in the health care experiment, reminding respondents of the argument of the dissenting Justices (Reminder 3) entirely eliminated the influence of the Court that we saw in all the other subgroups (No Reminder, Reminder 1, and Reminder 2). We believe that the argument offered by the dissent that government should not be able to force Americans to buy a product was a strong frame because this was the main justification offered by mandate opponents. 16 Thus, these patterns are consistent with Hypothesis 2 and with Chong and Druckman s earlier research. In the immigration experiment, each of our reminders had a positive effect. This is likely because, unlike in the health care experiment, many people did not receive news of the Court decision or misunderstood the main holding due to the confusing coverage. As we would expect, the coefficient on Reminder 2 (holding + argument in favor) is larger than the coefficient of either of the two other reminders, but the difference is not statistically significant. In the immigration experiment, the Court came down unanimously on the law s central provision, and thus we were not able to test for the effect of a dissent. However, we were able to emphasize in Reminders 2 and 3 that the decision was unanimous. This could drive the somewhat larger effects of these treatments, relative to Reminder 1, which only provides the Court s upholding of the law. Hypothesis 3 predicts that various individual characteristics mediate responsiveness to the Court. Table A1 investigates these patterns for respondents who received no reminder from us. Table A2 includes all respondents in the health care experiment, and Table A3 includes all respondents in the immigration experiment. For all regressions we report OLS coefficients, to facilitate the interpretation of coefficients, especially because we include interaction effects (Ai and Norton 2003). Results are very similar when we run ordered logit models, as Table A4 shows. An important mediator is trust in the Court; prior research suggests that credible interlocutors are more persuasive. We classified people who reported a high degree of confidence in the Court in Wave 1 as having high trust in the court, and persons reporting only some degree of confidence or hardly any confidence as having low trust. In both the immigration and health care experiments, people who correctly understood the decision (labeled aware ) and expressed high trust in the 16 More specifically, in January 2012, the Kaiser poll reported that 30 percent of opponents of the mandate cited government overreach as their main concern. Another 25 percent thought insurance would be too expense, a further 22 percent complained about the fine, while five percent opposed the mandate because they thought it was unconstitutional (Kaiser 2012c). 12

14 Court at Time 1 were most likely to change their opinions and follow the Court s lead at Time 2, controlling for a variety of other characteristics in the sample that received no reminder from us (Table A1). In the full sample, we see a statistically significant effect of trust on opinion change only in the immigration experiment. In our regression analyses, we focus on trust in Time 1 only, as this measure is independent of the Court s eventual decisions. We saw large shifts in trust following the health care case, because this decision was praised by Democrats and criticized by Republicans. We only saw modest shifts in trust following the immigration decision. An interesting question is whether people who changed their opinions to follow the Court s lead were more likely than others to maintain their trust in the Court. Our preliminary investigations suggest that this is the case. Table A5 suggests that groups that maintained or gained a high level of trust in the Court were more likely to follow the Court s lead than other groups. That said, this table also suggests that these groups may be very different from one another in terms of partisanship and other characteristics, so we set aside these findings for further examination. 17 Returning to our full sample, Table A2 analyzes the health care experiment. Model 1 compares people who received each of the treatments (Reminders 1-3) to people in the control (no reminder) group. Model 2 includes Reminder 3 (majority + dissent arguments) alone. Models 3 and 4 include these reminders plus a variety of other variables. Model 5 includes only those variables that are ever significant. Model 6 only includes people who received reminders from us. Our main findings are stable across specifications. Figure 2 below graphs how much each variable contributes to opinion change, based on our final specification for all respondents (Table A2, Model 5) FIGURE 2 HERE Consistent with our earlier descriptive results, the regression models confirm that people who received information about the holding of the decision from the news (labeled aware ) were more likely to change their minds and follow the lead of the Court. Only one of our experimental manipulations reminding people of the argument of the dissent had a significant (negative) impact on opinion change. Further analysis of Reminder 3 suggests that the dissent was powerful across ideologies; that is, in the group randomly assigned to this reminder, Democrats, Republicans, and Independents all kept their original views. 18 These findings suggest that, while experimenters can manipulate real- life effects by making certain counter- frames salient to respondents, simply repeating information many respondents already knew, as in Reminders 1 and 2, does not shift attitudes. This is consistent with the findings of Chong and Druckman (2007: 647), who also find that repetition does not have an impact on attitude change, even for low- information subjects. Taken together, these findings suggest our result is driven by processes of information transmission and is not an artifact of the experiment itself. 17 Additional analyses of shifts in trust are available upon request. 18 This analysis is available upon request. 13

15 Partisanship was an important mediator of responsiveness to the Court. Our expectation was that Independents would often follow the Court s lead because they have more malleable opinions, and this expectation was confirmed. We also found that Democrats responded more strongly to the Court s health care decision than did Republicans. This is not surprising, given that Democratic leaders hailed the decision, while Republican leaders criticized it. We also found that people with very low levels of education responded much less than did everyone else. This could mean that people with very low levels of education process information from the Court differently from more educated people, and do not connect the Court s decision about the constitutionality of the law to their policy views. It is also possible to see this effect because our measures of knowledge of the decision and interest in the news imperfectly captured all the variation in information about the decision. We found that media viewership was important. Fox News viewers kept their original views, while non- Fox viewers moved towards the Court. Fox covered the health care case extensively, with coverage both critical of the ruling and emphasizing the dissent. This lack of change among Fox viewers is similar in direction and magnitude to the experimental effect of Reminder 3: People we exposed to the dissent, like people Fox exposed to the dissent, were not swayed by the Court s decision. We found that Democrats, Republicans and Independents exposed to the dissent all kept their original views; similarly, more detailed analysis shows that Fox viewers with diverse party affiliations all kept their original views. That said, the Fox effect could also be selection, if, for example, people chose to watch Fox because they had firm views on health care. The immigration experiment also reflects findings that are consistent with our theoretical expectations and with the health care experiment results. In Table A3, Model 1 includes the randomly assigned reminders alone. Models 2, 3, 4, and 5 include these reminders plus a variety of other variables. Our main specification includes only those variables that are ever significant; these results are presented graphically in Figure 3 below. 19 Model 7 includes only people who we exposed to information about the case. 19 A somewhat unusual feature of our final specification of the immigration results (Table A3, Model 6, graphed in Figure 3 and replicated using ordered logit in Table A4) is that it includes a dummy variable labeled aware & high trust that takes on the value of 1 when people were aware of the decision through the media and had high trust in the Court, and 0 otherwise. A cursory glance might suggest that we are inappropriately including an interaction term without its constituent parts (Brambor et al. 2006). Models 4 and 5 in Table A3 present alternative specifications; Model 4 separately includes awareness of the decision, and trust in the Court, which could be thought of as the constituent parts of this interaction terms. Model 5 breaks down people into four groups, and compares three of them (people with high awareness and high trust, people with high awareness and low trust, and people with low awareness and high trust) against the omitted category of low awareness and low trust. Results across these specifications are highly consistent and suggest that movement is concentrated only among the group that has both high awareness 14

16 FIGURE 3 HERE Each of our experimental reminders about the Court decision increased the number of people who followed the Court s lead, likely because coverage of the decision through the media was confusing. Consistent with our expectations, we found that people who both correctly identified the Court holding and had high levels trust in the Court at the time of the first survey responded more than did others. As in the health care experiment, people with very low education responded less than did other people, likely for the same reasons outlined above. As in the health care experiment, we found a particularly strong response among Independents. This suggests that Independents have flexible views and follow the Court after both conservative and liberal rulings. Democrats also responded more strongly than Republicans, though we saw positive shifts in both groups. The comparatively small positive shift among Republicans could result from a ceiling effect 85 percent of Republicans already supported the Arizona law s key provision even before the Court upheld it and 89 percent supported the law after the decision. We found that race and ethnicity mattered for the immigration experiment, but not for health care. Hispanics moved in the opposite direction of the immigration ruling from May to June. This effect is highly concentrated among people who watch Spanish- language news on Univision or Telemundo. It is very likely that, as with the Fox effect in the health care case, concentrated and critical coverage of the immigration case on Univision and Telemundo led viewers to move away from the Court s ruling. 20 This data supports Hypothesis 4: that citizens may move away from the Court s position if a decision generates much critical news coverage to which they are exposed. In contrast, we did not find race to be a significant predictor of Court- led changes in health care attitudes. Tesler (2012) finds a racial gap between black and white Americans on Obama s health care legislation, which he attributes to positive associations between Obama and the legislation among black Americans. The racial cues generated by Obama talking about health care may not be generated when Supreme Court delivers the message instead. While our data only measure short- term opinion shifts, because there has been consistent polling on the Affordable Care Act over time, we can also examine whether the shift we observe are temporary or long lasting. Figure 4 below tracks and high trust in the Court. Following Brambor et al. (2006: 68-69), we only omit the constituent terms because we have strong theoretical reasons to do so (we do not expect shifts among people who trust the Court but aren t aware of its decisions), and because we included the full specification with the constituent terms and found their coefficients to be indistinguishable from A selection story is possible in the case of Spanish language television, but less plausible than in the case of Fox. That is, we would have to posit that people who had malleable views, and in particular people whose support for the Arizona law was particularly tenuous, chose to watch Univision and Telemundo. 15