Can Elites Shape Public Attitudes Toward Immigrants?: Evidence from the 2016 US Presidential Election

Can Elites Shape Public Attitudes Toward Immigrants?: Evidence from the 2016 US Presidential... Abstract It is well known that political elites can shape public attitudes toward policies and values. Less is known, however, about whether elites can also influence public perceptions of social groups they praise or denounce. I test this by analyzing the attitudinal effects of Donald Trump’s 2016 presidential campaign announcement speech, in which he referred to Mexican immigrants as “rapists” and “criminals.” First, to provide causal estimates, I analyze survey data using a counterfactual approach. I find evidence that Trump’s statements negatively affected public opinion toward immigrants particularly among groups with restrictionist tendencies. Second, using a panel survey experiment, I corroborate this causal relationship but find that these effects are short-lived. This explains why restrictionist politicians like Trump constantly prod natives to keep their messages’ effects from dissipating. I also find that only negative messages are consequential and find no evidence that elite statements are more impactful than those from non-elites, suggesting that the power of elite rhetoric lies primarily in its capacity to reach the masses via the news media. Introduction It is well known that elites can shape public attitudes toward policies, politicians, and belief statements (Mendelberg 2001). Less is known, however, about whether political elites can also influence public perceptions of the social groups they praise or denounce. Scholars have argued that politicians’ use of symbolic language blaming vulnerable groups, such as racial minorities, immigrants, and poor families, for society’s problems may not only encourage popular support for exclusionary policies but also influence public views of these groups themselves (Beckett 1997; Calavita 1996; Chavez 2008; Citrin, Reingold, and Green 1990; Edelman 1977; Santa Ana 2002). Though this hypothesis has been applied to the study of multiple policy domains, including welfare, crime/policing, immigration, and family (Beckett 1997; Chavez 2001; Flores 2015; Soss and Schram 2007), there is relatively little direct evidence that elites’ statements do in fact shape public attitudes toward targeted groups. In this paper, I test this widespread assumption by assessing the effect of anti-immigrant rhetoric by political elites on public attitudes toward immigrants as a case study of this long-held hypothesis. Assessing the effect of politicians’ rhetoric on immigration is not an easy task. Some of this complexity stems from the fact that immigration goes through cycles of low and high politicization (Massey and Pren 2012). In periods of high politicization, it becomes difficult to assess the power of a specific politician’s statements due to the large number of public figures opining about immigration simultaneously. Further, politicians’ statements may be endogenous to public mood about migration. Nevertheless, Donald Trump’s 2016 presidential campaign may provide an empirical opportunity to do so. On June 16, 2015, Trump announced his presidential candidacy in a speech in which he explicitly referred to Mexican immigrants as “rapists” and “criminals.” This speech received substantial media coverage (Burke 2016; Confessore and Yourish 2016). Since, up until that point, no other presidential candidate during the 2016 election cycle had made immigration a central campaign topic1 (Sanneh 2015), Trump’s speech provides an ideal empirical opportunity to assess whether politicians’ statements about immigrants do in fact shape immigration attitudes. I use two distinct methodologies that have unique strengths to address this question. First, I use a Gallup survey that was on the field during Trump’s speech and I compare the immigration attitudes of respondents interviewed right before the speech with those interviewed right after. Because the survey’s day of application was random, it allows me to approximate a natural experiment by treating respondents as if they were randomly assigned to pre- and post-treatment groups. I find evidence that Trump’s negative statements negatively affected public opinion toward immigrants particularly among Republicans and individuals without college degrees. While this survey analysis addresses concerns about external validity and realism by design, since the sample used is nationally representative and it examines a real life event, it cannot disentangle whether the effect was driven by Trump’s original speech or by subsequent media coverage or whether it led to permanent or temporary attitudinal changes. To address these questions, I conduct a national panel survey experiment in which I follow respondents that were exposed to both pro- and anti-immigrant messages over several weeks to test the duration of these messages’ attitudinal effects. Not only does the panel survey experiment corroborate the causal link between politicians’ statements and immigration attitudes, but it also uncovers three structural traits. First, these effects are ephemeral. They dissipate within days. Second, these effects are circumscribed to negative statements, as positive ones do not seem consequential. Third, they are not author-dependent. Statements by politicians were not more impactful than statements by local residents. This suggests that the power of elite rhetoric primarily lies in its capacity to reach the masses via the news media. Prior scholars have argued that elite statements that are explicitly racial in nature, like Trump’s speech on immigration, may fail to have attitudinal effects given widely held social norms against the open expression of racial prejudice (Mendelberg 2001). Nevertheless, my findings imply that immigration statements may be an exception. Immigrants’ contested legality may allow critics to explicitly target them without seemingly violating anti-racist social norms. Political Elites and Public Opinion There is ample evidence that political elites can influence public opinion. Researchers have found that elites can affect public attitudes toward laws (Druckman 2001; Mendelberg 2001; Nicholson 2011, 2012); values and statements (Goren, Federico, and Kittilson 2009; Kuklinski and Hurley 1994); political candidates (Arceneaux 2008); and even medical practices (Joslyn and Haider-Markel 2006). We know less, however, about elites’ capacity to influence public opinion about social groups they extol or condemn. Public opinion toward policies may be different from attitudes toward social groups. For example, both conservative and liberal elites typically describe “immigration laws” as “broken” (Massey and Sánchez 2010). In contrast, since national ideology defines the United States as a “nation of immigrants,” mainstream elites tend to portray “immigrants” in a positive light and highlight their contributions (Alba and Nee 2009). The interdisciplinary “symbolic politics” perspective suggests that elite statements may shape public views of targeted groups. From this perspective, individuals acquire affective predispositions such as ethnocentrism, racial attitudes, and altruism through socialization early in life. These predispositions guide their attitudes toward social and political issues (Easton and Dennis 1969; Hyman 1959; Sears 1993). Politicians employ symbols, images, and metaphors to evoke these predispositions and mobilize public sentiment about specific issues (Beckett 1997). When promoting divisive public policies, politicians use symbolic language that implicitly identifies social groups, such as racial minorities, immigrants, and poor families, as the source of social ailments (Beckett 1997; Calavita 1996; Edelman 1977). When these symbolic appeals connect with people’s emotional predispositions, the general public will often rally around punitive policies that target specific subgroups (Sears 1993). In addition to encouraging public support for exclusionary policies, scholars have claimed that such symbolic political discourse may also shape public views of the targeted groups (Calavita 1996; Chavez 2001, 2008; Santa Ana 2002). However, previous studies have not firmly established whether restrictionist politicians are causing or merely echoing public opinion. It is certainly plausible that politicians’ statements could have an independent effect on public opinion toward minorities, but it is equally possible that they merely echo the views of the general population by adopting anti-minority stances. A related literature has examined the impact of “frames” used by elites on public attitudes in a more empirical fashion2 (Nelson and Kinder 1996). These studies define framing as “alternative conceptualizations of an issue or event” and have found that, under certain conditions, frames could shape public opinion (Druckman and Nelson 2003). For example, referring to anti-poverty programs as “welfare” reduces the public’s approval of the programs, but framing those same programs as “helping the poor” increases popular support, since the latter label taps into an altruistic predisposition among the public. Nevertheless, though frames commonly enter political discourse via political actors like politicians or political parties, most framing studies have relied on newspaper vignettes or unattributed statements to expose individuals to different frames (Druckman, Peterson, and Slothuus 2013). Some studies have assessed the impact of political parties, most often via party endorsements, on public attitudes (Kam 2005; Slothuus 2010). This literature suggests that political parties may be most effective at shaping their own members’ attitudes. However, it is not entirely clear that this finding extends also to individual politicians who, as individuals, may or may not be seen as true representatives of their entire parties. While scholars theorize that the statements of public figures like politicians or judges may be more consequential than the words of common people (Edelman 1977; Sunstein 1996), evidence about this is mixed (Joslyn and Haider-Markel 2006; Kuklinski and Hurley 1994). This raises questions about whether political statements are always consequential regardless of their author or source. The characteristics of the message may also matter. Mendelberg (2001) examines the effect of political statements on individuals’ policy positions on race, welfare, poverty, spending, and defense. She finds that political messages that are only implicitly racial are more effective in shaping public opinion than those in which race is explicitly mentioned. She argues that elite statements that make “direct verbal references to race” will backfire since they violate the norm of equality. Other authors have similarly claimed that in the present time period, there is a strong social norm against the open expression of racial or ethnic prejudice (Blinder, Ford, and Ivarsflaten 2013; Bonilla-Silva 2003). Therefore, contemporary US politicians will instead use covert or coded language when blaming minorities for societal problems in an attempt to mobilize the majority population. According to this perspective, if an elite statement is obviously racial, as in Donald Trump’s statement, we should expect that this social norm will become activated and neutralize the statements’ effects since most people do not want to be seen as prejudiced (Mendelberg 2001). Public Opinion Toward Immigration Understanding public attitudes toward immigrants is a key area of social science research (Bloemraad, Silva, and Voss 2016; Marrow 2011; Schachter 2015). This vast literature has been generally oriented toward two camps. A materialist camp, with deep roots in economics, argues that the way natives respond to immigrants is deeply colored by their own individual material self-interest (Hainmueller and Hopkins 2014). A second approach, pursued by sociologists and political scientists, finds that public opinion toward immigration is rooted in its perceived cultural and economic impacts on the nation as a whole (Citrin, Reingold, and Green 1990; Hopkins, Tran, and Williamson 2014). This study is not designed to adjudicate between these two competing literatures. Rather, it aims to provide causal evidence for one of the key mechanisms through which cultural or economic concerns presumably become activated in the minds of the general people: elites’ public statements. A significant number of studies assume, but do not directly test, that politicians have the power to shape public attitudes by constructing narratives through which immigrants are viewed (Hopkins 2010; Massey and Sánchez 2010; Santa Ana 2002). Politicians are assumed to be “external, politicizing agents” that have the power to “politicize people’s day-to-day experiences” and even make “people’s views turn anti-immigrant” (Hopkins 2010). Though there is ample evidence that politicians can elect to portray immigrants in positive or negative ways (Brown 2013; Massey and Pren 2012; Santa Ana 2002), there is less direct evidence that their statements do in fact influence public opinion, which I test in this paper. In a previous paper, I adopted a counterfactual approach to examine the effect of SB 1070, a high-profile anti-immigrant law passed by Arizona in 2010, on both public attitudes and behaviors toward immigrants using Twitter data from that state. I found that SB 1070 had a negative impact on the sentiment of the average tweet regarding immigrants, Mexicans, and Hispanics, but not on tweets about Asians or Blacks (Flores 2017). Though that paper demonstrated that restrictionist policies could set into motion attitudinal dynamics, I could not determine whether the effects were driven by the tough language used by anti-immigrant politicians in Arizona or by the passage of the policy itself, which may have legitimated extreme views. In this paper, I address this gap by employing a causal inference approach to examine whether politicians’ statements by themselves can have an independent effect on attitudes toward immigrants. Immigration and the 2016 Presidential Election On June 16, 2015, Donald Trump held a campaign rally in New York City to announce he was running for president of the United States of America. Though Trump was primarily known as a businessman and television personality, he had been personally involved in politics by endorsing candidates, donating money, and publicly promoting his policy preferences for decades (Haberman and Burns 2016). He had also run for office multiple times starting in 2000, when he first campaigned for the US presidency and won two Reform Party primaries. Trump had a history of making racially charged public statements, such as his accusation that President Barack Obama was not born in the United States (Krieg 2016). However, when it came to immigration, his ideological position had constantly shifted. In August 2013, he had actually declared his support for amnesty for undocumented immigrants (Bandler 2016). Up until that point, no other presidential candidates in the 2016 campaign had made immigration a central issue (Sanneh 2015). This was partly the result of a concerted effort by the Republican leadership to avoid taking overly punitive stances on immigration (Haberman 2016; Republican National Committee 2013). Such efforts were conceived by Republican leaders after Mitt Romney lost to Barack Obama and only obtained 27 percent of the Hispanic vote in the 2012 presidential election. These leaders believed that Romney’s hard-line position on immigration and especially his call for the “self-deportation” of millions of undocumented immigrants had alienated Hispanic voters (Franke-Ruta 2013; Rove 2013; Rubin 2013). Nevertheless, Trump broke with the party line by bringing immigration into the forefront in his announcement speech. Standing in front of journalists from all over the country, he said: “When Mexico sends its people, they’re not sending their best. They’re not sending you. They’re not sending you. They’re sending people that have lots of problems, and they’re bringing those problems with us. They’re bringing drugs. They’re bringing crime. They’re rapists. And some, I assume, are good people. But I speak to border guards and they tell us what we’re getting. And it only makes common sense. It only makes common sense. They’re sending us not the right people. It’s coming from more than Mexico. It’s coming from all over South and Latin America, and it’s coming probably—probably—from the Middle East. But we don’t know. Because we have no protection and we have no competence, we don’t know what’s happening. And it’s got to stop and it’s got to stop fast” (Trump 2015). Trump’s speech, which was widely covered by the media, proved influential. Eventually, all the other Republican candidates, with the exception of Jeb Bush, took a hard-line approach to immigration (Ehrenfreund 2016). Since no other prominent politician was talking about immigration prior to his speech, Trump’s announcement provides an ideal opportunity for social scientists to study the consequences of politicians’ statements on immigration attitudes. Since he was the only prominent politician publicly taking a hard-line position on immigration, his statements could be considered an independent shock on public opinion toward immigration. If multiple politicians were making similar statements concomitantly, it would be hard to determine whether any of them was single-handedly affecting public sentiment. Journalists and other observers have claimed that Trump’s speech was deeply consequential. Critics have claimed that it energized white supremacists and hardened public attitudes toward immigrants (Burke 2016; Carroll 2016; Haberman 2016). Immigrants and even US-born Hispanics have reported increased public harassment following Trump’s speech, which includes strangers asking them if they speak English or inquiring about their citizenship status (Vasquez 2015). Further, actual violence has been directly linked to this speech. In August 2015, two white men beat with a pole and urinated on a 58-year-old homeless Mexican man in Boston. The men cited Trump as the reason why they did it. “Trump was right” about “deporting all these illegals,” they said (Fox 2015; Vasquez 2015). Nevertheless, beyond this anecdotal evidence, no systematic study has been conducted to examine the social consequences of Trump’s pronouncements. After all, these incidents could be isolated cases. In addition, there are theoretical reasons why this speech may not have been socially consequential. As mentioned earlier, scholars have suggested that when elite statements are openly racial, like in Trump’s case, attitudinal effects may not occur. This is so because such overt racial language may activate the social norm against the open expression of racial prejudice among individuals. In this paper, I employ two distinct methodologies and use Trump’s announcement speech as a quasi-natural experiment to systematically assess whether elites’ pronouncements on immigration can shape public attitudes toward immigrants. In addition, to address aforementioned literature gaps, I test whether these effects are temporary or durable and whether the identity of the speaker matters. Data and Methods As mentioned earlier, Donald Trump gave his presidential announcement speech on June 16, 2015. By happenstance, Gallup was conducting a nationwide survey on minority rights and relations during that week. The survey included the most commonly used question to gauge attitudes toward immigrants: “Thinking now about immigrants—that is, people who come from other countries to live here in the United States. In your view, should immigration be kept at its present level, increased, or decreased?” The availability of this data and, most importantly, its collection timing, allow me to conduct a counterfactual analysis similar to a regression discontinuity design in which the date of the interview was randomly assigned to respondents (Legewie 2013). By comparing the responses of survey takers that were interviewed right before Trump’s speech with those that Gallup reached slightly after, I can estimate the effect of Donald Trump’s announcement on public attitudes toward immigrants. However, for such comparison to provide quasi-causal estimates, I must show that there was no selection into treatment. In other words, aside from the timing of the interview, both groups of respondents did not differ in any meaningful way that could have affected their immigration views. Gallup began polling on June 15 and ended on July 10. Three hundred fifty-eight respondents were interviewed on June 15 and 16. These respondents form the control group in my study. Trump’s speech took place on June 16, but most media outlets covered his speech on June 17. Indeed, figure 1 shows that mentions of Donald Trump in the seven largest US newspapers increased substantially only on June 17. Therefore, I consider people interviewed on June 15 and 16 to be a good indicator of the prevailing mood about immigration in the United States before Trump’s speech.3 A total of 769 individuals were interviewed in the two following days (June 17–18). This group constitutes the treated group since they were likely exposed to Trump’s rhetoric. For this design to be valid, it needs to meet several conditions. First, there has to be a discontinuity in the treatment. In this case, I need to show that after Trump’s speech, popular media picked up his speech and, hence, individuals in the treated group were more likely to be exposed to more negative portrayals of immigrants. Figure 1. View largeDownload slide Newspaper mentions of Donald Trump (June 2015) Note: The blue line shows the total number of articles mentioning Donald Trump published by the seven largest newspapers in the United States: the New York Times, Los Angeles Times, Washington Post, Wall Street Journal, Boston Globe, USA Today, and Orange County Register. Vertical line indicates the day when Donald Trump launched his presidential campaign (June 16, 2015). The horizontal scale, ranging from 9 to 23, indicates the day of the month. Figure 1. View largeDownload slide Newspaper mentions of Donald Trump (June 2015) Note: The blue line shows the total number of articles mentioning Donald Trump published by the seven largest newspapers in the United States: the New York Times, Los Angeles Times, Washington Post, Wall Street Journal, Boston Globe, USA Today, and Orange County Register. Vertical line indicates the day when Donald Trump launched his presidential campaign (June 16, 2015). The horizontal scale, ranging from 9 to 23, indicates the day of the month. According to the literature, for a politician’s statement to be socially consequential, it has to receive media coverage. To test that, I examine media coverage from the seven largest newspapers in the United States: the New York Times, Los Angeles Times, Washington Post, Wall Street Journal, Boston Globe, USA Today, and Orange County Register. These newspapers cover a wide range of the mainstream ideological spectrum. In addition, given that some individuals may not read newspapers but instead get their news from radio, TV, or Internet sources, I also analyze transcripts from all radio shows produced by the non-partisan National Public Radio (NPR) as well as transcripts from TV news and online articles produced by the Cable News Corporation (CNN).4 Figure 2 shows the number of news items related to immigration produced by each of these media outlets. The vertical line indicates when Donald Trump launched his presidential campaign. O, in this case, represents the week when Trump gave his speech. The horizontal scale, ranging from −4 to +4, indicates the number of weeks since Trump’s announcement. Figure 2 indicates that the absolute number of news items related to immigration increased in six out of nine news outlets after Trump’s announcement. This increase of media coverage is statistically significant at the 99 percent level. The emotional language used by Trump in referring to immigrants may have contributed to the rapid spread of his message (Bail 2015).5 Figure 2. View largeDownload slide Media coverage of immigration (May 19–July 13, 2015)Note: Lines show the total number of articles related to immigration published by each of the seven largest newspapers in the United States: the New York Times, Los Angeles Times, Washington Post, Wall Street Journal, Boston Globe, USA Today, and Orange County Register. It also includes news coverage data from all radio shows produced by National Public Radio (NPR) as well as transcripts from TV news and online articles produced by the Cable News Corporation (CNN). Vertical line indicates when Donald Trump launched his presidential campaign. The horizontal scale, ranging from −4 to +4, indicates the number of weeks since Trump’s announcement. The bold red line labeled “crime + immig” indicates the number of immigration-related articles published in any of the seven newspapers that mentioned rape, drugs, or crime. Figure 2. View largeDownload slide Media coverage of immigration (May 19–July 13, 2015)Note: Lines show the total number of articles related to immigration published by each of the seven largest newspapers in the United States: the New York Times, Los Angeles Times, Washington Post, Wall Street Journal, Boston Globe, USA Today, and Orange County Register. It also includes news coverage data from all radio shows produced by National Public Radio (NPR) as well as transcripts from TV news and online articles produced by the Cable News Corporation (CNN). Vertical line indicates when Donald Trump launched his presidential campaign. The horizontal scale, ranging from −4 to +4, indicates the number of weeks since Trump’s announcement. The bold red line labeled “crime + immig” indicates the number of immigration-related articles published in any of the seven newspapers that mentioned rape, drugs, or crime. Further, I also find that the content of these news items changed by examining the articles published in these seven national newspapers. The red line indicates the number of all immigration-related articles in these newspapers that mentioned rape, drugs, or crime. It shows that a week before Trump’s speech, the number of newspaper articles linking immigrants to crime, which was the centerpiece of this politician’s announcement, doubled, going from 29 to 63. Such figure was again 63 one week later, and it increased to 143 and 202 articles two and three weeks after Trump’s announcement, respectively. This suggests that there could be two components to the treatment: the initial Trump speech as well as the subsequent media coverage and discussion triggered by the speech, which I will address in the second part of the article. The second condition I need to meet is that assignment to the treatment and control groups has to be random. Or, perhaps more realistically, there should be no major socio-demographic differences between both groups. If there were large compositional differences between these two groups, my design would be invalid, since these differences could be driving any perceived changes in attitudes. I use a relatively narrow window of time since there were plenty of respondents interviewed during those four days, but also because increasing the window period by adding more days would probably result in unobserved heterogeneity in the sample. In other words, survey takers interviewed on later dates may be increasingly different than the control group. However, in subsequent analyses I show the robustness of my results to the use of a wider time period. Gallup Survey Results Table 1 shows the descriptive statistics of the Gallup Minority Rights’ Survey for the analyzed period (June 15–18, 2015). This time period captures respondents interviewed two days after Trump’s speech and two days before. I use two-tailed two-sample t-tests assuming unequal variances to test whether there are any significant socio-demographic differences between respondents interviewed before Trump’s announcement and those interviewed after Trump’s announcement. I test a number of individual characteristics that could shape immigration attitudes, including age, gender, race, education, partisanship, community size, employment status, and geographic region. Table 1 shows that there are almost no statistically significant differences in the sample composition before and after Trump’s speech. Both samples are quite comparable in terms of their race, ethnicity, age, employment status, community type, and partisanship. The only exception is living in the Midwest or Western regions. The sample contains more Midwestern and fewer Western residents in the post-Trump period. I will return to this later on. In addition, a Gallup manager told me that the team in charge of this survey produced a sampling frame every day based on a national random sample of land- and cell-phone lines from across the country, which they call a “replicate.” This explains why there are no substantial socio-demographic differences in the sample composition by day, since every day Gallup composed a random national snapshot of US residents, weighted by key socio-demographic weights. This information corroborates that interviewing order was designed to be random. Table 1. Gallup Minority Rights’ Survey: Descriptive Statistics (n = 1,104) Mean/% Before Trump After Trump Race  White 64.5 64.8 64.3  African American 19.6 17.3 20.8  Other 15.7 17.8 14.8 Female 44.3 41.6 45.6 Hispanic 18.1 20.3 17.1 Age categories  18–29 23.1 22.8 23.4  30–49 40.3 40.1 40.4  50–64 56.6 56.9 56.6  65+ 74.1 74.1 74.1 Rural/small town 26.7 26.5 26.9 Unemployed 6.6 8.1 6.0 College 29.4 28.5 29.9 Geographic area  East 16.6 18.1 15.8  Midwest 16.5 11.0 19.2*  South 26.6 25.4 27.1  West 40.1 45.5 37.7* Partisanship  Democrat 37.1 36.0 37.7  Independent 39.4 43.2 37.7  Republican 20.8 18.4 21.9 Mean/% Before Trump After Trump Race  White 64.5 64.8 64.3  African American 19.6 17.3 20.8  Other 15.7 17.8 14.8 Female 44.3 41.6 45.6 Hispanic 18.1 20.3 17.1 Age categories  18–29 23.1 22.8 23.4  30–49 40.3 40.1 40.4  50–64 56.6 56.9 56.6  65+ 74.1 74.1 74.1 Rural/small town 26.7 26.5 26.9 Unemployed 6.6 8.1 6.0 College 29.4 28.5 29.9 Geographic area  East 16.6 18.1 15.8  Midwest 16.5 11.0 19.2*  South 26.6 25.4 27.1  West 40.1 45.5 37.7* Partisanship  Democrat 37.1 36.0 37.7  Independent 39.4 43.2 37.7  Republican 20.8 18.4 21.9 Source: Gallup Minority Rights’ Survey 2015. Descriptive statistics include respondents interviewed between June 15 and 18, 2015. “After Trump” indicates that the survey interview was conducted on either June 17 or 18 (after Donald Trump’s presidential campaign announcement). Two-tailed two-sample t-tests assuming unequal variances compare differences between respondents interviewed before and those reached after Trump’s announcement. *** p < 0.001 ** p < 0.01 * p < 0.05. Table 1. Gallup Minority Rights’ Survey: Descriptive Statistics (n = 1,104) Mean/% Before Trump After Trump Race  White 64.5 64.8 64.3  African American 19.6 17.3 20.8  Other 15.7 17.8 14.8 Female 44.3 41.6 45.6 Hispanic 18.1 20.3 17.1 Age categories  18–29 23.1 22.8 23.4  30–49 40.3 40.1 40.4  50–64 56.6 56.9 56.6  65+ 74.1 74.1 74.1 Rural/small town 26.7 26.5 26.9 Unemployed 6.6 8.1 6.0 College 29.4 28.5 29.9 Geographic area  East 16.6 18.1 15.8  Midwest 16.5 11.0 19.2*  South 26.6 25.4 27.1  West 40.1 45.5 37.7* Partisanship  Democrat 37.1 36.0 37.7  Independent 39.4 43.2 37.7  Republican 20.8 18.4 21.9 Mean/% Before Trump After Trump Race  White 64.5 64.8 64.3  African American 19.6 17.3 20.8  Other 15.7 17.8 14.8 Female 44.3 41.6 45.6 Hispanic 18.1 20.3 17.1 Age categories  18–29 23.1 22.8 23.4  30–49 40.3 40.1 40.4  50–64 56.6 56.9 56.6  65+ 74.1 74.1 74.1 Rural/small town 26.7 26.5 26.9 Unemployed 6.6 8.1 6.0 College 29.4 28.5 29.9 Geographic area  East 16.6 18.1 15.8  Midwest 16.5 11.0 19.2*  South 26.6 25.4 27.1  West 40.1 45.5 37.7* Partisanship  Democrat 37.1 36.0 37.7  Independent 39.4 43.2 37.7  Republican 20.8 18.4 21.9 Source: Gallup Minority Rights’ Survey 2015. Descriptive statistics include respondents interviewed between June 15 and 18, 2015. “After Trump” indicates that the survey interview was conducted on either June 17 or 18 (after Donald Trump’s presidential campaign announcement). Two-tailed two-sample t-tests assuming unequal variances compare differences between respondents interviewed before and those reached after Trump’s announcement. *** p < 0.001 ** p < 0.01 * p < 0.05. Having largely ruled out differences in sample composition, I now examine whether Trump’s speech affected public attitudes toward immigrants. As mentioned earlier, Gallup included the most commonly used question to measure immigration attitudes, which reads: “In your view, should immigration be kept at its present level, increased, or decreased?” In figure 3, I plot the responses to our main question by days since Trump’s speech. The vertical red line indicates day 0, which is when Trump gave his official speech. Day 1 represents the day after this speech, which is when most media outlets began reacting to his pronouncement. Figure 3. View largeDownload slide Individual preferences over immigration (N = 2,004) Source: Gallup’s Survey on Minority Rights and Relations 2015. Question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Vertical line indicates the day of Donald Trump’s presidential campaign announcement (June 16, 2015). Figure 3. View largeDownload slide Individual preferences over immigration (N = 2,004) Source: Gallup’s Survey on Minority Rights and Relations 2015. Question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Vertical line indicates the day of Donald Trump’s presidential campaign announcement (June 16, 2015). Figure 3 shows that public opinion on immigrants hardened after Trump’s speech. While 33 percent of respondents believed that immigration levels should be decreased a day after Trump’s speech, only 25 percent believed so a day before. Similarly, 23 percent of survey takers stated that immigration flows should be increased a day after the speech. Such number was 36 percent the day before. Figure 3 also shows a gradual hardening trend over this time period. For example, 10 days after Trump’s speech, 39 percent of respondents believed immigration levels should be decreased, which represents a 56 percent increase in this group relative to the day when Trump gave his speech. I now turn to regression analysis to examine these results in a more systematic manner. Since respondents could choose among three different options to express their opinion about current immigration flows (i.e., “decrease,” “increase,” and “keep the same”), I use multinomial regression models to predict their responses. The reference category for all models is “increase.” The key coefficient is AfterTrump, which indicates being interviewed after Trump’s speech. Therefore, this coefficient represents the effect of this speech on the individual’s immigration attitudes. Table 2 shows the results in relative risk ratios. All models include a dummy variable that indicates whether respondents were interviewed after Trump’s speech. In model 1, I compare respondents that were interviewed a day after Trump’s speech with those interviewed one day before. It shows that respondents interviewed after Trump’s speech had a higher relative risk to choose either “decrease” or “keep the same” relative to choosing “increase” than those interviewed before.6 These differences are statistically significant at the 99 percent level. In model 2, I test the robustness of this finding to expanding the time window to two days before and two days after the speech. While choosing “same” is no longer significantly different relative to choosing “increase,” respondents were still more likely to favor decreasing immigration flows even after expanding the time window. I consider this model to be the best specification since it maximizes the number of respondents in both control and treatment conditions but also keeps the time window relatively narrow, which minimizes concerns about selection (since respondent heterogeneity could increase as the time window expands). Table 2. Regression Models Predicting Individual Support for Immigration (Gallup Survey) Model 1: One day Model 2: Two days Model 3: Four days Model 4: Seven days Model 5: Is immigration bad? VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.802** 2.022** 1.279 1.407* 1.328+ 1.416* 1.244 1.425* 1.598* (0.396) (0.483) (0.200) (0.234) (0.202) (0.230) (0.177) (0.216) (0.365) Constant 1.052 0.701* 1.211 0.895 1.211 0.895 1.211 0.895 0.211*** (0.168) (0.125) (0.153) (0.122) (0.153) (0.122) (0.153) (0.122) (0.038) Observations 501 501 1,104 1,104 1,253 1,253 1,855 1,855 493 Model 1: One day Model 2: Two days Model 3: Four days Model 4: Seven days Model 5: Is immigration bad? VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.802** 2.022** 1.279 1.407* 1.328+ 1.416* 1.244 1.425* 1.598* (0.396) (0.483) (0.200) (0.234) (0.202) (0.230) (0.177) (0.216) (0.365) Constant 1.052 0.701* 1.211 0.895 1.211 0.895 1.211 0.895 0.211*** (0.168) (0.125) (0.153) (0.122) (0.153) (0.122) (0.153) (0.122) (0.038) Observations 501 501 1,104 1,104 1,253 1,253 1,855 1,855 493 Source: Gallup Minority Rights’ Survey 2015. Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 +p < 0.10. “After Trump” indicates that the survey interview was conducted after Donald Trump’s presidential campaign announcement. Analyzed question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Models 1–4 show results, in relative risk ratios, for multinomial logistic regression predicting responses to this question. Reference category for these multinomial regression models is “increased.” Model 1 includes all respondents interviewed the day of the speech and the day after. Model 2 includes all respondents interviewed two days before and two days after. Model 3 includes respondents interviewed two days before and four days after. Model 4 includes all respondents interviewed two days before and seven days later. The last column, model 5, analyzes the following question: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” This logistic regression predicts responding “bad thing” relative to responding “good thing” or “mixed.” Results for model 5 shown in odd ratios. Table 2. Regression Models Predicting Individual Support for Immigration (Gallup Survey) Model 1: One day Model 2: Two days Model 3: Four days Model 4: Seven days Model 5: Is immigration bad? VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.802** 2.022** 1.279 1.407* 1.328+ 1.416* 1.244 1.425* 1.598* (0.396) (0.483) (0.200) (0.234) (0.202) (0.230) (0.177) (0.216) (0.365) Constant 1.052 0.701* 1.211 0.895 1.211 0.895 1.211 0.895 0.211*** (0.168) (0.125) (0.153) (0.122) (0.153) (0.122) (0.153) (0.122) (0.038) Observations 501 501 1,104 1,104 1,253 1,253 1,855 1,855 493 Model 1: One day Model 2: Two days Model 3: Four days Model 4: Seven days Model 5: Is immigration bad? VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.802** 2.022** 1.279 1.407* 1.328+ 1.416* 1.244 1.425* 1.598* (0.396) (0.483) (0.200) (0.234) (0.202) (0.230) (0.177) (0.216) (0.365) Constant 1.052 0.701* 1.211 0.895 1.211 0.895 1.211 0.895 0.211*** (0.168) (0.125) (0.153) (0.122) (0.153) (0.122) (0.153) (0.122) (0.038) Observations 501 501 1,104 1,104 1,253 1,253 1,855 1,855 493 Source: Gallup Minority Rights’ Survey 2015. Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 +p < 0.10. “After Trump” indicates that the survey interview was conducted after Donald Trump’s presidential campaign announcement. Analyzed question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Models 1–4 show results, in relative risk ratios, for multinomial logistic regression predicting responses to this question. Reference category for these multinomial regression models is “increased.” Model 1 includes all respondents interviewed the day of the speech and the day after. Model 2 includes all respondents interviewed two days before and two days after. Model 3 includes respondents interviewed two days before and four days after. Model 4 includes all respondents interviewed two days before and seven days later. The last column, model 5, analyzes the following question: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” This logistic regression predicts responding “bad thing” relative to responding “good thing” or “mixed.” Results for model 5 shown in odd ratios. Model 3 is similar, but I expand the window of comparison to two days before and four days after the speech. I find that respondents interviewed after the speech were still more likely to choose “decrease” than those surveyed a day earlier. Similarly, model 4 shows that the result is robust to increasing the time window to respondents interviewed within a seven-day period after Trump’s speech. The fact that an effect can still be found even after increasing the time window may indicate that Trump’s speech may have resulted in a relatively durable hardening of public opinion toward immigrants. Alternatively, it could also mean that there was an ongoing hardening trend of public opinion during this time period, perhaps motivated by ongoing media coverage and discussion of Trump’s speech. I address this point using an original survey experiment in the second part of the article. If Trump’s speech hardened public opinion toward immigrants, we should expect other immigration-related questions in the same survey to be similarly affected. Fortunately, Gallup included another question on immigration in the same survey. This question read: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” Respondents could choose “bad thing,” “good thing,” or “mixed.” I use a logistic regression to analyze this question. This model predicts responding “bad thing” relative to responding “good thing” or “mixed.” I combined these two last categories because only nine respondents chose “mixed” during this time window. In model 5, shown in table 3, I compare respondents that were interviewed a day after Trump’s speech with those Gallup reached a day before. In accordance with our previous results, individuals interviewed right after Trump’s speech were more likely to view immigration as a “bad thing” than those interviewed earlier. Table 3. Multinomial Logistic Regressions Predicting Individual Support for Immigration (Gallup Survey) Model 1: Region Model 2: Demog Model 3: SES Model 4: Full VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.239 1.365+ 1.234 1.337+ 1.250 1.361+ 1.275 1.361+ (0.195) (0.230) (0.198) (0.235) (0.203) (0.245) (0.210) (0.251) Region (ref = East)  Midwest 1.369 1.515 1.227 1.306 1.219 1.162 1.290 1.257 (0.352) (0.431) (0.321) (0.379) (0.324) (0.349) (0.345) (0.390)  South 1.000 2.016** 0.989 2.051** 0.954 1.911* 0.982 1.893* (0.234) (0.502) (0.234) (0.522) (0.229) (0.497) (0.240) (0.511)  West 0.937 1.116 0.971 1.162 0.962 1.111 1.006 1.188 (0.194) (0.260) (0.207) (0.283) (0.209) (0.275) (0.226) (0.308)  Female 0.942 1.026 0.977 1.043 1.017 1.136 (0.145) (0.167) (0.153) (0.174) (0.162) (0.196)  Hispanic 0.670* 0.640* 0.643* 0.583* 0.645* 0.622* (0.133) (0.140) (0.132) (0.132) (0.137) (0.146) Race (ref = White)  Black 0.684+ 0.639* 0.680+ 0.603* 0.790 1.001 (0.134) (0.137) (0.137) (0.133) (0.171) (0.241)  Other 0.707 0.806 0.738 0.821 0.778 0.951 (0.152) (0.187) (0.163) (0.196) (0.176) (0.242) Age 1.007 1.021*** 1.006 1.022*** 1.008+ 1.024*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.005) Education (ref = less than HS)  HS 1.345 1.183 1.450 1.358 (0.569) (0.497) (0.620) (0.623)  Technical/trade 0.374 1.193 0.448 1.826 (0.271) (0.696) (0.334) (1.202)  Incomplete college 1.556 1.437 1.612 1.632 (0.650) (0.595) (0.683) (0.736)  Complete college 0.824 0.620 0.878 0.769 (0.347) (0.262) (0.376) (0.351)  Graduate school 0.891 0.478+ 0.921 0.583 (0.377) (0.206) (0.396) (0.270) Unemployed 0.501* 0.752 0.503* 0.781 (0.151) (0.236) (0.155) (0.254) Partisanship (ref = Republican)  Democrat 0.581* 0.226*** (0.140) (0.057)  Independent 1.049 0.694 (0.249) (0.168)  Other 0.993 0.996 (0.589) (0.599) Community type (ref = big city)  Small city 0.873 0.903 (0.206) (0.238)  Big city suburb 0.907 0.739 (0.201) (0.189)  Small city suburb 1.165 1.167 (0.427) (0.482)  Town 0.862 1.042 (0.243) (0.321)  Rural 0.844 1.044 (0.250) (0.317) Constant 1.208 0.673+ 1.125 0.282*** 1.078 0.329* 1.164 0.448 (0.239) (0.156) (0.358) (0.098) (0.553) (0.172) (0.664) (0.263) Observations 1,104 1,104 1,091 1,091 1,080 1,080 1,080 1,080 Pseudo R-squared 0.010 0.010 0.029 0.029 0.046 0.046 0.070 0.070 Model 1: Region Model 2: Demog Model 3: SES Model 4: Full VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.239 1.365+ 1.234 1.337+ 1.250 1.361+ 1.275 1.361+ (0.195) (0.230) (0.198) (0.235) (0.203) (0.245) (0.210) (0.251) Region (ref = East)  Midwest 1.369 1.515 1.227 1.306 1.219 1.162 1.290 1.257 (0.352) (0.431) (0.321) (0.379) (0.324) (0.349) (0.345) (0.390)  South 1.000 2.016** 0.989 2.051** 0.954 1.911* 0.982 1.893* (0.234) (0.502) (0.234) (0.522) (0.229) (0.497) (0.240) (0.511)  West 0.937 1.116 0.971 1.162 0.962 1.111 1.006 1.188 (0.194) (0.260) (0.207) (0.283) (0.209) (0.275) (0.226) (0.308)  Female 0.942 1.026 0.977 1.043 1.017 1.136 (0.145) (0.167) (0.153) (0.174) (0.162) (0.196)  Hispanic 0.670* 0.640* 0.643* 0.583* 0.645* 0.622* (0.133) (0.140) (0.132) (0.132) (0.137) (0.146) Race (ref = White)  Black 0.684+ 0.639* 0.680+ 0.603* 0.790 1.001 (0.134) (0.137) (0.137) (0.133) (0.171) (0.241)  Other 0.707 0.806 0.738 0.821 0.778 0.951 (0.152) (0.187) (0.163) (0.196) (0.176) (0.242) Age 1.007 1.021*** 1.006 1.022*** 1.008+ 1.024*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.005) Education (ref = less than HS)  HS 1.345 1.183 1.450 1.358 (0.569) (0.497) (0.620) (0.623)  Technical/trade 0.374 1.193 0.448 1.826 (0.271) (0.696) (0.334) (1.202)  Incomplete college 1.556 1.437 1.612 1.632 (0.650) (0.595) (0.683) (0.736)  Complete college 0.824 0.620 0.878 0.769 (0.347) (0.262) (0.376) (0.351)  Graduate school 0.891 0.478+ 0.921 0.583 (0.377) (0.206) (0.396) (0.270) Unemployed 0.501* 0.752 0.503* 0.781 (0.151) (0.236) (0.155) (0.254) Partisanship (ref = Republican)  Democrat 0.581* 0.226*** (0.140) (0.057)  Independent 1.049 0.694 (0.249) (0.168)  Other 0.993 0.996 (0.589) (0.599) Community type (ref = big city)  Small city 0.873 0.903 (0.206) (0.238)  Big city suburb 0.907 0.739 (0.201) (0.189)  Small city suburb 1.165 1.167 (0.427) (0.482)  Town 0.862 1.042 (0.243) (0.321)  Rural 0.844 1.044 (0.250) (0.317) Constant 1.208 0.673+ 1.125 0.282*** 1.078 0.329* 1.164 0.448 (0.239) (0.156) (0.358) (0.098) (0.553) (0.172) (0.664) (0.263) Observations 1,104 1,104 1,091 1,091 1,080 1,080 1,080 1,080 Pseudo R-squared 0.010 0.010 0.029 0.029 0.046 0.046 0.070 0.070 Source: Gallup Minority Rights’ Survey 2015. Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 +p < 0.10. “After Trump” indicates that the survey interview was conducted after Donald Trump’s presidential campaign announcement. Respondent sample for all models consists of individuals interviewed two days before Trump’s announcement and two days after. Analyzed question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Models show results for multinomial logistic regression predicting responses to this question. Reference category for all models is “increased.” Table 3. Multinomial Logistic Regressions Predicting Individual Support for Immigration (Gallup Survey) Model 1: Region Model 2: Demog Model 3: SES Model 4: Full VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.239 1.365+ 1.234 1.337+ 1.250 1.361+ 1.275 1.361+ (0.195) (0.230) (0.198) (0.235) (0.203) (0.245) (0.210) (0.251) Region (ref = East)  Midwest 1.369 1.515 1.227 1.306 1.219 1.162 1.290 1.257 (0.352) (0.431) (0.321) (0.379) (0.324) (0.349) (0.345) (0.390)  South 1.000 2.016** 0.989 2.051** 0.954 1.911* 0.982 1.893* (0.234) (0.502) (0.234) (0.522) (0.229) (0.497) (0.240) (0.511)  West 0.937 1.116 0.971 1.162 0.962 1.111 1.006 1.188 (0.194) (0.260) (0.207) (0.283) (0.209) (0.275) (0.226) (0.308)  Female 0.942 1.026 0.977 1.043 1.017 1.136 (0.145) (0.167) (0.153) (0.174) (0.162) (0.196)  Hispanic 0.670* 0.640* 0.643* 0.583* 0.645* 0.622* (0.133) (0.140) (0.132) (0.132) (0.137) (0.146) Race (ref = White)  Black 0.684+ 0.639* 0.680+ 0.603* 0.790 1.001 (0.134) (0.137) (0.137) (0.133) (0.171) (0.241)  Other 0.707 0.806 0.738 0.821 0.778 0.951 (0.152) (0.187) (0.163) (0.196) (0.176) (0.242) Age 1.007 1.021*** 1.006 1.022*** 1.008+ 1.024*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.005) Education (ref = less than HS)  HS 1.345 1.183 1.450 1.358 (0.569) (0.497) (0.620) (0.623)  Technical/trade 0.374 1.193 0.448 1.826 (0.271) (0.696) (0.334) (1.202)  Incomplete college 1.556 1.437 1.612 1.632 (0.650) (0.595) (0.683) (0.736)  Complete college 0.824 0.620 0.878 0.769 (0.347) (0.262) (0.376) (0.351)  Graduate school 0.891 0.478+ 0.921 0.583 (0.377) (0.206) (0.396) (0.270) Unemployed 0.501* 0.752 0.503* 0.781 (0.151) (0.236) (0.155) (0.254) Partisanship (ref = Republican)  Democrat 0.581* 0.226*** (0.140) (0.057)  Independent 1.049 0.694 (0.249) (0.168)  Other 0.993 0.996 (0.589) (0.599) Community type (ref = big city)  Small city 0.873 0.903 (0.206) (0.238)  Big city suburb 0.907 0.739 (0.201) (0.189)  Small city suburb 1.165 1.167 (0.427) (0.482)  Town 0.862 1.042 (0.243) (0.321)  Rural 0.844 1.044 (0.250) (0.317) Constant 1.208 0.673+ 1.125 0.282*** 1.078 0.329* 1.164 0.448 (0.239) (0.156) (0.358) (0.098) (0.553) (0.172) (0.664) (0.263) Observations 1,104 1,104 1,091 1,091 1,080 1,080 1,080 1,080 Pseudo R-squared 0.010 0.010 0.029 0.029 0.046 0.046 0.070 0.070 Model 1: Region Model 2: Demog Model 3: SES Model 4: Full VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.239 1.365+ 1.234 1.337+ 1.250 1.361+ 1.275 1.361+ (0.195) (0.230) (0.198) (0.235) (0.203) (0.245) (0.210) (0.251) Region (ref = East)  Midwest 1.369 1.515 1.227 1.306 1.219 1.162 1.290 1.257 (0.352) (0.431) (0.321) (0.379) (0.324) (0.349) (0.345) (0.390)  South 1.000 2.016** 0.989 2.051** 0.954 1.911* 0.982 1.893* (0.234) (0.502) (0.234) (0.522) (0.229) (0.497) (0.240) (0.511)  West 0.937 1.116 0.971 1.162 0.962 1.111 1.006 1.188 (0.194) (0.260) (0.207) (0.283) (0.209) (0.275) (0.226) (0.308)  Female 0.942 1.026 0.977 1.043 1.017 1.136 (0.145) (0.167) (0.153) (0.174) (0.162) (0.196)  Hispanic 0.670* 0.640* 0.643* 0.583* 0.645* 0.622* (0.133) (0.140) (0.132) (0.132) (0.137) (0.146) Race (ref = White)  Black 0.684+ 0.639* 0.680+ 0.603* 0.790 1.001 (0.134) (0.137) (0.137) (0.133) (0.171) (0.241)  Other 0.707 0.806 0.738 0.821 0.778 0.951 (0.152) (0.187) (0.163) (0.196) (0.176) (0.242) Age 1.007 1.021*** 1.006 1.022*** 1.008+ 1.024*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.005) Education (ref = less than HS)  HS 1.345 1.183 1.450 1.358 (0.569) (0.497) (0.620) (0.623)  Technical/trade 0.374 1.193 0.448 1.826 (0.271) (0.696) (0.334) (1.202)  Incomplete college 1.556 1.437 1.612 1.632 (0.650) (0.595) (0.683) (0.736)  Complete college 0.824 0.620 0.878 0.769 (0.347) (0.262) (0.376) (0.351)  Graduate school 0.891 0.478+ 0.921 0.583 (0.377) (0.206) (0.396) (0.270) Unemployed 0.501* 0.752 0.503* 0.781 (0.151) (0.236) (0.155) (0.254) Partisanship (ref = Republican)  Democrat 0.581* 0.226*** (0.140) (0.057)  Independent 1.049 0.694 (0.249) (0.168)  Other 0.993 0.996 (0.589) (0.599) Community type (ref = big city)  Small city 0.873 0.903 (0.206) (0.238)  Big city suburb 0.907 0.739 (0.201) (0.189)  Small city suburb 1.165 1.167 (0.427) (0.482)  Town 0.862 1.042 (0.243) (0.321)  Rural 0.844 1.044 (0.250) (0.317) Constant 1.208 0.673+ 1.125 0.282*** 1.078 0.329* 1.164 0.448 (0.239) (0.156) (0.358) (0.098) (0.553) (0.172) (0.664) (0.263) Observations 1,104 1,104 1,091 1,091 1,080 1,080 1,080 1,080 Pseudo R-squared 0.010 0.010 0.029 0.029 0.046 0.046 0.070 0.070 Source: Gallup Minority Rights’ Survey 2015. Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 +p < 0.10. “After Trump” indicates that the survey interview was conducted after Donald Trump’s presidential campaign announcement. Respondent sample for all models consists of individuals interviewed two days before Trump’s announcement and two days after. Analyzed question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Models show results for multinomial logistic regression predicting responses to this question. Reference category for all models is “increased.” Did the effect of Trump’s speech vary within certain subpopulations? It is plausible that this speech had different effects for several reasons. First, perhaps certain subgroups like Republicans or working-class respondents were more likely to watch the presidential announcement of a Republican candidate like Donald Trump. Second, as predicted by the framing literature (Druckman, Peterson, and Slothuus 2013), Trump’s message could have resonated more among partisan groups: individuals predisposed to holding restrictionist views on immigration. To test this, in models not shown, I explore subgroup differences in the effect of Donald Trump’s speech. Figure 4 shows the results of these models in graphic form. It shows the effect of Trump’s speech on the predicted probabilities of wanting to decrease immigration flows by respondents’ education, partisanship, and race. In accordance with prior framing studies, it shows that Republican respondents and those without a college education were more likely to choose “decrease” after Trump’s speech at statistically significant levels. In contrast, the responses of college-educated, Democrat, and independent voters did not change significantly. Non-Hispanic whites were also more likely to express a desire to reduce immigration after Trump’s announcement, though such difference was within the margin of error. This evidence indicates that the attitudinal effects of Trump’s speech were largely driven by attitudinal changes among groups that would seem especially receptive to his message including lower-educated and Republican respondents, who tend to have restrictionist views on immigration. Figure 4. View largeDownload slide Predicted probabilities of wanting to decrease immigration flows Source: Gallup’s Survey on Minority Rights and Relations 2015. Graph shows the predicted probabilities of wanting to “decrease” immigration flows based on multinomial logistic regression models predicting responses to this question: “In your view, should immigration be kept at its present level, increased, or decreased?” Reference category for these multinomial regression models is “kept same.” “After” indicates that the survey interview was conducted after Donald Trump’s presidential campaign announcement. Sample includes all respondents interviewed two days after the speech and two days before. Figure 4. View largeDownload slide Predicted probabilities of wanting to decrease immigration flows Source: Gallup’s Survey on Minority Rights and Relations 2015. Graph shows the predicted probabilities of wanting to “decrease” immigration flows based on multinomial logistic regression models predicting responses to this question: “In your view, should immigration be kept at its present level, increased, or decreased?” Reference category for these multinomial regression models is “kept same.” “After” indicates that the survey interview was conducted after Donald Trump’s presidential campaign announcement. Sample includes all respondents interviewed two days after the speech and two days before. Robustness checks As I showed earlier, individuals interviewed before Trump’s speech did not differ much from those interviewed after (see table 1). The one exception is that Midwest residents were more likely and Western residents less likely to be interviewed after Trump’s presidential announcement. Therefore, in this section I test whether results are robust to the inclusion of socio-demographic controls, including geographic residence. The regression form is identical to the first equation, except that I add a vector of individual characteristics including age, race, ethnicity, sex, region, partisanship, community type, education, and employment status. Table 3 reproduces the multinomial logistic regression model predicting attitudes toward immigration flows. The respondent sample for all models consists of individuals interviewed within two days before Trump’s announcement and two days after. Model 1 includes controls for geographic residence. It shows that even after including these controls, being interviewed after Trump’s speech is correlated with wanting to decrease immigration flows (though this is significant only at the 90 percent level). Model 2 tests the robustness of these results to adding gender, race, age, and ethnicity indicators. Despite these controls, results remain positive and statistically significant. In model 3, I include controls for educational attainment and employment. Even with these controls, the association between being interviewed after Trump’s speech and wanting to decrease immigration flows remains positive and statistically significant. Finally, in model 4, I include all controls from model 3 but also add controls for partisanship and community type. Results remain robust to this model specification as well. Effects on Other Policies To further test the robustness of these findings, I analyze responses to another set of policies: affirmative action policies targeting women and minorities. If responses to these policies also varied after Trump’s announcement, this could be evidence of selection bias in the sense that certain respondents, such as conservative individuals, could have been more likely to be interviewed after Trump’s speech. Respondents were asked if they favored or opposed these programs for women and for “racial minorities” separately. I use a logistic regression to predict opposition to these programs. I test the effect of Trump’s speech on these items using different time intervals: one day before and one day after the speech, two days before and two days after, and finally, two days before and four days after. Table 4 shows that respondents interviewed after Trump’s speech did not have more conservative views on these items than those interviewed before it. Therefore, these findings further assuage concerns about selection bias. Table 4. Logistic Regressions Predicting Opposition to Affirmative Action Policies M1: 1 day M2: 2 days M3: 4 days M4: 1 day M5: 2 days M6: 4 days VARIABLES Women Women Women Minorities Minorities Minorities After Trump 1.012 0.947 0.860 1.007 0.929 0.857 (0.204) (0.135) (0.120) (0.189) (0.126) (0.114) Constant 0.421*** 0.435*** 0.435*** 0.656** 0.593*** 0.593*** (0.064) (0.051) (0.051) (0.093) (0.066) (0.066) Observations 484 1,083 1,228 486 1,079 1,223 M1: 1 day M2: 2 days M3: 4 days M4: 1 day M5: 2 days M6: 4 days VARIABLES Women Women Women Minorities Minorities Minorities After Trump 1.012 0.947 0.860 1.007 0.929 0.857 (0.204) (0.135) (0.120) (0.189) (0.126) (0.114) Constant 0.421*** 0.435*** 0.435*** 0.656** 0.593*** 0.593*** (0.064) (0.051) (0.051) (0.093) (0.066) (0.066) Observations 484 1,083 1,228 486 1,079 1,223 Source: Gallup Minority Rights’ Survey 2015. Odd ratios reported. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 +p < 0.10. Question read: “Do you generally favor or oppose affirmative action programs for women?” Second question was identical except that it asked about “racial minorities.” Answers were recoded so that 1 indicated opposition and 0 support for these policies. Models 1 and 4 include all respondents interviewed the day of Trump’s speech and the day after. Models 2 and 5 include all respondents interviewed two days before and two days after. Models 3 and 6 include respondents interviewed two days before and four days after. Table 4. Logistic Regressions Predicting Opposition to Affirmative Action Policies M1: 1 day M2: 2 days M3: 4 days M4: 1 day M5: 2 days M6: 4 days VARIABLES Women Women Women Minorities Minorities Minorities After Trump 1.012 0.947 0.860 1.007 0.929 0.857 (0.204) (0.135) (0.120) (0.189) (0.126) (0.114) Constant 0.421*** 0.435*** 0.435*** 0.656** 0.593*** 0.593*** (0.064) (0.051) (0.051) (0.093) (0.066) (0.066) Observations 484 1,083 1,228 486 1,079 1,223 M1: 1 day M2: 2 days M3: 4 days M4: 1 day M5: 2 days M6: 4 days VARIABLES Women Women Women Minorities Minorities Minorities After Trump 1.012 0.947 0.860 1.007 0.929 0.857 (0.204) (0.135) (0.120) (0.189) (0.126) (0.114) Constant 0.421*** 0.435*** 0.435*** 0.656** 0.593*** 0.593*** (0.064) (0.051) (0.051) (0.093) (0.066) (0.066) Observations 484 1,083 1,228 486 1,079 1,223 Source: Gallup Minority Rights’ Survey 2015. Odd ratios reported. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 +p < 0.10. Question read: “Do you generally favor or oppose affirmative action programs for women?” Second question was identical except that it asked about “racial minorities.” Answers were recoded so that 1 indicated opposition and 0 support for these policies. Models 1 and 4 include all respondents interviewed the day of Trump’s speech and the day after. Models 2 and 5 include all respondents interviewed two days before and two days after. Models 3 and 6 include respondents interviewed two days before and four days after. More substantively, these findings also disprove the notion of spillover effects of negative messages toward immigrants. One hypothesis could be that being exposed to these negative messages may make respondents more punitive in general regardless of the policy under consideration. Nevertheless, this evidence suggests that immigration statements primarily affect attitudes toward immigrants. The lack of effect on affirmative action policies may stem from the fact that respondents may see these laws as primarily affecting African Americans. Taken together, these results provide suggestive evidence that Donald Trump’s public speech portraying immigrants as rapists and criminals shaped public attitudes toward immigrants. This research design has multiple strengths. First, these results are robust to multiple robustness checks. In addition, since the setup is based on a real-life political event, it is highly realistic. Finally, the data set used is nationally representative, which assuages concerns about external validity. Nevertheless, the research design also has some limitations. First, figure 3 shows that after Trump’s speech, there was a gradual hardening of public opinion toward immigrants during the month of July. How can we explain this? Did Trump’s speech result in a permanent shift of attitudes? An alternative explanation is that Trump’s announcement had only a short-term effect but its impact was extended and perhaps intensified by ongoing media coverage of his speech. Media coverage has been shown to shape public opinion (Terkildsen and Schnell 1997). After all, as I show in figure 2, media attention to immigration increased and also became more negative after Trump’s speech. Unfortunately, this research design cannot disentangle these two processes. In addition, this analysis is based on the plausible but untested assumption that individuals were exposed to Trump’s speech, but this may not necessarily be the case (Barabas and Jerit 2010). Further, even if Trump’s speech was consequential, we are uncertain whether it was impactful because a recognized political actor gave it as part of a presidential campaign or whether the message would have been equally impactful had it been uttered by anyone else given the same level of media attention. Finally, since Trump’s statement toward immigrants was negative, we are unable to examine the counterfactual: whether a positive message would have been equally consequential. Panel Survey Experiment To address these lingering questions, I conducted an original national panel survey experiment in which individuals were randomly exposed to different immigration-related messages uttered by both politicians and non-politicians. In this experiment, respondents were asked to read a brief newspaper article containing a statement regarding immigration, which was experimentally manipulated to vary in its overall sentiment and in its author. Further, to test whether messages’ effects are durable or transitory, respondents were re-interviewed a few weeks after initially completing the survey. This experiment addresses several limitations of the initial analysis based on Gallup survey data. First, by relying on random assignment, it provides unbiased causal estimates of the power of politicians’ messages to shape public attitudes toward immigrants. Second, by testing whether the author of the message (politician or non-politician) is consequential, it is able to directly test the main hypothesis and disentangle the effects of messages sent by politicians from other messages that individuals may be exposed to in their daily lives. Third, since the panel experiment re-interviews respondents a few weeks after the initial interview, it is able to test whether the effects of being exposed to immigration messages are durable or transitory, which is a common critique of standard one-shot experimental research designs. The survey experiment, which was approved by the Institutional Review Board of the University of Michigan (HUM00113743), was implemented in spring 2016. Before I began collecting data, I registered the research design and pre-analysis plan at Evidence in Governance and Politics (EGAP), an online repository of social science experiments and observational studies (ID 20160318AA). In all, 1,430 adult US residents were recruited via Amazon’s Mechanical Turk Internet service, which is a website where users perform tasks like responding to online surveys for small payments. Ninety-two percent of respondents, or 1,317 individuals, finished the survey. Average survey completion time was four minutes. Each participant was paid 32 cents. The sample is not strictly representative of the US population. However, table 5 shows that it provides significantly more variation in terms of race, ethnicity, gender, age, geography, and socio-economic background than traditional experimental studies conducted on college students. Further, recent evidence suggests that national samples of online respondents provide very similar results to samples that are explicitly designed to be representative (Weinberg, Freese, and McElhattan 2014). Table 5. Survey Experiment: Descriptive Statistics (n = 1,317) Mean/% (SD) Min Max Race/ethnicity  White 75.17  African American 6.45  Asian 7.29  Hispanic 7.90  Other 3.19 Age 34.37 (11.19) 18 75 Community size 2.18 (0.91) 1 4 Education 4.1 (1.33) 1 8 State  California 11.0  Florida 7.0  Texas 7.0  New York 6.0  Pennsylvania 6.0  Ohio 5.0  Michigan 4.0  Illinois 4.0  North Carolina 3.0  Georgia 3.0  New Jersey 3.0  Indiana 3.0  Virginia 2.0  Washington 3.0  Other States 34.0 Mean/% (SD) Min Max Race/ethnicity  White 75.17  African American 6.45  Asian 7.29  Hispanic 7.90  Other 3.19 Age 34.37 (11.19) 18 75 Community size 2.18 (0.91) 1 4 Education 4.1 (1.33) 1 8 State  California 11.0  Florida 7.0  Texas 7.0  New York 6.0  Pennsylvania 6.0  Ohio 5.0  Michigan 4.0  Illinois 4.0  North Carolina 3.0  Georgia 3.0  New Jersey 3.0  Indiana 3.0  Virginia 2.0  Washington 3.0  Other States 34.0 Note: Community size 1 = large, 2 = medium-size city, 3 = small town, and 4 = rural. The education categories are 1 = less than HS, 2 = HS, 3 = some college, 4 = 2-year college, 5 = 4-year college, 6 = master’s, 7 = professional degree (i.e., MBA, JD, etc.), and 8 = PhD. Table 5. Survey Experiment: Descriptive Statistics (n = 1,317) Mean/% (SD) Min Max Race/ethnicity  White 75.17  African American 6.45  Asian 7.29  Hispanic 7.90  Other 3.19 Age 34.37 (11.19) 18 75 Community size 2.18 (0.91) 1 4 Education 4.1 (1.33) 1 8 State  California 11.0  Florida 7.0  Texas 7.0  New York 6.0  Pennsylvania 6.0  Ohio 5.0  Michigan 4.0  Illinois 4.0  North Carolina 3.0  Georgia 3.0  New Jersey 3.0  Indiana 3.0  Virginia 2.0  Washington 3.0  Other States 34.0 Mean/% (SD) Min Max Race/ethnicity  White 75.17  African American 6.45  Asian 7.29  Hispanic 7.90  Other 3.19 Age 34.37 (11.19) 18 75 Community size 2.18 (0.91) 1 4 Education 4.1 (1.33) 1 8 State  California 11.0  Florida 7.0  Texas 7.0  New York 6.0  Pennsylvania 6.0  Ohio 5.0  Michigan 4.0  Illinois 4.0  North Carolina 3.0  Georgia 3.0  New Jersey 3.0  Indiana 3.0  Virginia 2.0  Washington 3.0  Other States 34.0 Note: Community size 1 = large, 2 = medium-size city, 3 = small town, and 4 = rural. The education categories are 1 = less than HS, 2 = HS, 3 = some college, 4 = 2-year college, 5 = 4-year college, 6 = master’s, 7 = professional degree (i.e., MBA, JD, etc.), and 8 = PhD. Respondents were asked to read a brief newspaper article containing a statement on immigration. The author (politician or non-politician) and the statements’ overall sentiment (neutral, pro-, or anti-immigrant) were randomly manipulated. The experiment had a 3 × 2 between-subjects design presented in table 6. There were four treatment conditions plus a control: (1) positive statement about immigrants by local resident; (2) positive statement about immigrants by politician; (3) negative statement about immigrants by local resident; (4) negative statement about immigrants by politician; and (5) neutral statement about immigrants (control condition: no speaker identified). Table 6. Experimental Design (n = 1,317) Immigration stance Pro-immigrant Anti-immigrant Politician Condition #1 (270) Condition #2 (256) Non-politician Condition #3 (276) Condition #4 (256) Immigration stance Pro-immigrant Anti-immigrant Politician Condition #1 (270) Condition #2 (256) Non-politician Condition #3 (276) Condition #4 (256) Note: Table shows the number of respondents on each experimental condition. A fifth condition, the control group, had 259 respondents. Table 6. Experimental Design (n = 1,317) Immigration stance Pro-immigrant Anti-immigrant Politician Condition #1 (270) Condition #2 (256) Non-politician Condition #3 (276) Condition #4 (256) Immigration stance Pro-immigrant Anti-immigrant Politician Condition #1 (270) Condition #2 (256) Non-politician Condition #3 (276) Condition #4 (256) Note: Table shows the number of respondents on each experimental condition. A fifth condition, the control group, had 259 respondents. To increase the realism of the treatment, the message was constructed based on real statements uttered by several high-profile restrictionist politicians, including Donald Trump, Pennsylvania congressman Lou Barletta, and former Arizona governor Jan Brewer. To preclude the identification of any of these figures based on this message, their statements were thoroughly blended and reworded to preserve their spirit but without making their origin too obvious. Pretesting was done via Amazon’s Mechanical Turk to verify that respondents could not identify the author of the statement. In the message, the speaker lists what he perceives are the negative consequences of immigration, including crime and increased health care and education costs. In the positive version, the same message is modified to change the sentiment of the message while keeping its structure and length intact. For example, while the negative version of the message ends with “we cannot stand idly by as these criminals compromise our quality of life,” the positive one does so with “we should support these immigrants since they improve our quality of life.” For experimental reasons, the messages were designed to be mirror images of each other. However, it is possible that this reduced the positivity of the positive message, since it included the same references as the negative message to immigration-related costs even if it questioned them. Finally, a control condition was added in which immigration was primed in a neutral way. This condition described how the US Census defines immigration. All treatment conditions had a similar word length—between 87 and 96 words. See Appendix A for a full description of all conditions. Past research shows that messengers’ traits may affect the attitudinal effects of elite cues (Kuklinski and Hurley 1994). Controversial messengers like Dr. Kevorkian or the KKK may weaken the impact of the message on opinion (Joslyn and Haider-Markel 2006; Nicholson 2011). In this case, the controversial status of Donald Trump could have depressed the effect of its immigration message among liberals but magnified it among conservatives. To test this, the identity of the speaker was randomly manipulated to include a non-controversial politician and, to determine whether the messenger’s status matters, a non-politician local resident. Some respondents were told that Wyoming governor Matt Mead had uttered the message, and others were told that a “local resident” named Matt Reed had made it. Governor Mead, a Republican, was chosen because Wyoming is the least populous state in the United States.7 As such, relatively few people would know Governor Mead’s actual position on immigration. This would make his utterance of either negative or positive statements about migration more credible than if we had used a more popular governor whose positions were better known. A randomization check is included in Appendix A. In addition, to test whether the effects of politicians’ statements are durable or ephemeral, respondents were randomly assigned to one of two experimental conditions: those contacted to take a follow-up survey after two weeks and those contacted after four weeks. Six hundred thirty-eight respondents were asked to take the survey after two weeks, and 454 of them (or 71 percent) completed the follow-up questionnaire. Six hundred thirty-one individuals were asked to be re-interviewed after four weeks, and 421 individuals (66.6 percent) agreed to do so. Though there could be concerns that these attrition rates could bias results, I find that there are no systematic socio-demographic differences between individuals who agreed to take the follow-up survey and those who did not (see table A2 in the Appendix). The only significant difference is that older respondents were more likely to take the follow-up survey. In models not shown, I include age controls but results do not change substantively. The follow-up surveys did not contain any experimental treatments. Instead, to test whether the statements respondents read during the baseline survey were still consequential, the follow-up surveys included the same questions about immigration. To verify the validity of the experimental treatments, I conducted a series of manipulation checks after the application of the experiment.8 I found that the different experimental conditions successfully conveyed the treatment cues I intended. Experimental Results In the survey experiment, I included the exact same attitudinal question asked in the Gallup survey, which read: “In your view, should immigration be kept at its present level, increased, or decreased?” Mirroring the prior analysis, I use a multinomial logistic regression to test whether the experimental treatments affected individuals’ expressed desire to decrease immigration flows. In these models, shown in table 7, I include controls for pre-treatment covariates like race/ethnicity, sex, age, and partisanship. This should tighten the estimates of treatment effects without increasing any bias because treatment assignment is unrelated to these traits9 (Gelman and Hill 2006). Table 7. Regression Models Predicting Immigration Attitudes (Survey Experiment) Change immigration flows? Is immigration good? VARIABLES Same Decrease Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.809 0.883 (0.307) (0.378) (0.579) (0.168)  Local negative 1.195 2.255** 2.016* 1.127 (0.295) (0.631) (0.664) (0.213)  Politician positive 1.347 1.451 1.307 0.867 (0.317) (0.409) (0.440) (0.163)  Politician negative 1.098 2.455** 2.676** 1.360 (0.271) (0.682) (0.848) (0.263) Race/ethnicity  Non-Hispanic Black 1.499 1.363 1.649 1.696* (0.516) (0.530) (0.630) (0.440)  Asian 0.984 0.590 0.547 0.822 (0.275) (0.214) (0.252) (0.191)  Hispanic 1.504 0.908 0.839 0.844 (0.459) (0.323) (0.298) (0.196)  Other 0.516 0.439+ 0.386 0.980 (0.194) (0.218) (0.295) (0.322) Conservative/moderate 1.736* 6.679*** 3.649*** 2.616*** (0.434) (1.688) (0.780) (0.393) College 0.980 0.573** 0.488*** 0.603*** (0.156) (0.103) (0.0931) (0.0739) Female 0.865 1.066 1.232 1.246 (0.141) (0.194) (0.238) (0.153) Age 1.020* 1.031*** 1.002 0.999 (0.008) (0.009) (0.008) (0.005) Constant 1.337 0.339** 0.132*** 0.891 (0.429) (0.124) (0.0538) (0.221) Observations 1,317 1,317 1,317 1,317 Pseudo R-squared 0.078 0.078 0.044 0.044 Change immigration flows? Is immigration good? VARIABLES Same Decrease Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.809 0.883 (0.307) (0.378) (0.579) (0.168)  Local negative 1.195 2.255** 2.016* 1.127 (0.295) (0.631) (0.664) (0.213)  Politician positive 1.347 1.451 1.307 0.867 (0.317) (0.409) (0.440) (0.163)  Politician negative 1.098 2.455** 2.676** 1.360 (0.271) (0.682) (0.848) (0.263) Race/ethnicity  Non-Hispanic Black 1.499 1.363 1.649 1.696* (0.516) (0.530) (0.630) (0.440)  Asian 0.984 0.590 0.547 0.822 (0.275) (0.214) (0.252) (0.191)  Hispanic 1.504 0.908 0.839 0.844 (0.459) (0.323) (0.298) (0.196)  Other 0.516 0.439+ 0.386 0.980 (0.194) (0.218) (0.295) (0.322) Conservative/moderate 1.736* 6.679*** 3.649*** 2.616*** (0.434) (1.688) (0.780) (0.393) College 0.980 0.573** 0.488*** 0.603*** (0.156) (0.103) (0.0931) (0.0739) Female 0.865 1.066 1.232 1.246 (0.141) (0.194) (0.238) (0.153) Age 1.020* 1.031*** 1.002 0.999 (0.008) (0.009) (0.008) (0.005) Constant 1.337 0.339** 0.132*** 0.891 (0.429) (0.124) (0.0538) (0.221) Observations 1,317 1,317 1,317 1,317 Pseudo R-squared 0.078 0.078 0.044 0.044 Note: Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. For the first model, respondents were asked: “In your view, should immigration be kept at its present level, increased, or decreased?” Reference category was “increased.” In the second question, respondents were asked: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” Response categories were: “good thing,” “bad thing,” and “mixed.” The reference category in the regression model was “good.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. Table 7. Regression Models Predicting Immigration Attitudes (Survey Experiment) Change immigration flows? Is immigration good? VARIABLES Same Decrease Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.809 0.883 (0.307) (0.378) (0.579) (0.168)  Local negative 1.195 2.255** 2.016* 1.127 (0.295) (0.631) (0.664) (0.213)  Politician positive 1.347 1.451 1.307 0.867 (0.317) (0.409) (0.440) (0.163)  Politician negative 1.098 2.455** 2.676** 1.360 (0.271) (0.682) (0.848) (0.263) Race/ethnicity  Non-Hispanic Black 1.499 1.363 1.649 1.696* (0.516) (0.530) (0.630) (0.440)  Asian 0.984 0.590 0.547 0.822 (0.275) (0.214) (0.252) (0.191)  Hispanic 1.504 0.908 0.839 0.844 (0.459) (0.323) (0.298) (0.196)  Other 0.516 0.439+ 0.386 0.980 (0.194) (0.218) (0.295) (0.322) Conservative/moderate 1.736* 6.679*** 3.649*** 2.616*** (0.434) (1.688) (0.780) (0.393) College 0.980 0.573** 0.488*** 0.603*** (0.156) (0.103) (0.0931) (0.0739) Female 0.865 1.066 1.232 1.246 (0.141) (0.194) (0.238) (0.153) Age 1.020* 1.031*** 1.002 0.999 (0.008) (0.009) (0.008) (0.005) Constant 1.337 0.339** 0.132*** 0.891 (0.429) (0.124) (0.0538) (0.221) Observations 1,317 1,317 1,317 1,317 Pseudo R-squared 0.078 0.078 0.044 0.044 Change immigration flows? Is immigration good? VARIABLES Same Decrease Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.809 0.883 (0.307) (0.378) (0.579) (0.168)  Local negative 1.195 2.255** 2.016* 1.127 (0.295) (0.631) (0.664) (0.213)  Politician positive 1.347 1.451 1.307 0.867 (0.317) (0.409) (0.440) (0.163)  Politician negative 1.098 2.455** 2.676** 1.360 (0.271) (0.682) (0.848) (0.263) Race/ethnicity  Non-Hispanic Black 1.499 1.363 1.649 1.696* (0.516) (0.530) (0.630) (0.440)  Asian 0.984 0.590 0.547 0.822 (0.275) (0.214) (0.252) (0.191)  Hispanic 1.504 0.908 0.839 0.844 (0.459) (0.323) (0.298) (0.196)  Other 0.516 0.439+ 0.386 0.980 (0.194) (0.218) (0.295) (0.322) Conservative/moderate 1.736* 6.679*** 3.649*** 2.616*** (0.434) (1.688) (0.780) (0.393) College 0.980 0.573** 0.488*** 0.603*** (0.156) (0.103) (0.0931) (0.0739) Female 0.865 1.066 1.232 1.246 (0.141) (0.194) (0.238) (0.153) Age 1.020* 1.031*** 1.002 0.999 (0.008) (0.009) (0.008) (0.005) Constant 1.337 0.339** 0.132*** 0.891 (0.429) (0.124) (0.0538) (0.221) Observations 1,317 1,317 1,317 1,317 Pseudo R-squared 0.078 0.078 0.044 0.044 Note: Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. For the first model, respondents were asked: “In your view, should immigration be kept at its present level, increased, or decreased?” Reference category was “increased.” In the second question, respondents were asked: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” Response categories were: “good thing,” “bad thing,” and “mixed.” The reference category in the regression model was “good.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. Table 7 shows that individuals who read the negative statement about immigrants were more likely to state that they would like to decrease the current immigration levels in the United States. While individuals who were assigned to the control condition had a 23 percent predicted probability to choose “decrease,” individuals who read the negative statement about immigrants uttered by a local resident or a politician had a 37 or 41 percent predicted probability of doing so, respectively. These findings go in the same direction as those found using the Gallup survey. Surprisingly, altering the author of the statement did not make much difference. What mattered the most was the polarity of the message, but even then only negative statements were consequential. Positive messages about immigrants had no effect on immigration attitudes regardless of whether they were uttered by a politician or a local resident. Mirroring the analysis in the previous section, I conduct interaction tests by education, partisanship, and race. Just like I found then, non-Hispanic whites, conservatives, and lower-educated respondents were more likely to express a desire to decrease immigration flows when primed with negative framings of immigration.10 For example, while 46 percent of non-Hispanic whites wanted to decrease immigration after reading the governor’s negative statement on immigrants, such figure is only 23 percent for non-whites in the same condition (and 24 percent for the control group). With regard to educational attainment, while 50 percent of non-college-educated respondents chose to decrease immigration when assigned to the negative/politician condition, only 33 percent of college-educated survey takers preferred to decrease immigration flows after reading the negative statement from the governor. Finally, conservative respondents also seemed more responsive to negative messages about immigration than liberal individuals.11 Seventy-four percent of conservative respondents wanted to decrease immigration flows when exposed to the negative/governor condition (relative to 46 percent when assigned to the control condition). In contrast, only 29 percent of their liberal counterparts also chose to reduce immigration when assigned to the same condition (and 18 percent in the control group). Were attitudinal effects of negative statements durable or ephemeral? I test this by re-interviewing some respondents after two weeks and others after four and asking them the same attitudinal questions as in the baseline survey. Figure 5 provides a summary of these findings. Full regression results can be found in the Appendix. As figure 5 shows, I find no significant effects after two weeks. Similarly, among those respondents re-interviewed after four weeks, there were no significant differences in the likelihood to choose “decrease” across the different experimental conditions (figure 5).12 Figure 5. View largeDownload slide Predicted probabilities of wanting to decrease current immigration levels Note: Predicted probabilities were estimated based on results from a multinomial logistic regression. The question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Response options were: “Present level,” “increased,” and “decreased.” The reference category in the regression model was “increased.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. “2 weeks” indicates respondents that took a follow-up survey two weeks after the baseline survey. “4 weeks” indicates respondents that took this follow-up survey four weeks after the original survey. Figure 5. View largeDownload slide Predicted probabilities of wanting to decrease current immigration levels Note: Predicted probabilities were estimated based on results from a multinomial logistic regression. The question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Response options were: “Present level,” “increased,” and “decreased.” The reference category in the regression model was “increased.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. “2 weeks” indicates respondents that took a follow-up survey two weeks after the baseline survey. “4 weeks” indicates respondents that took this follow-up survey four weeks after the original survey. To verify the robustness of these findings, I analyze the effect of immigration statements on a second attitudinal item also included in the Gallup survey. Respondents were asked: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” Response categories were: “good thing,” “bad thing,” and “mixed.” As table 8 shows, respondents exposed to negative messages about immigrants were more likely to claim that immigrants were a “bad thing” relative to those assigned to the neutral condition. The effect was slightly larger among those who read the negative statement by a politician (15 percent) than among those who were told the negative statement came from a local resident (12 percent). In contrast, only 7 percent of individuals in the baseline condition stated that immigrants were a “bad thing.” Just like I found with the first item, no statistically significant effect was detected among those exposed to positive statements. Table 8. Multinomial Logistic Regressions Predicting Immigration Attitudes (Survey Experiment) Baseline 2 Weeks 4 Weeks VARIABLES Bad Mixed Bad Mixed Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.809 0.883 1.301 1.321 1.170 1.059 (0.579) (0.168) (0.760) (0.447) (0.646) (0.351)  Local negative 2.016* 1.127 1.670 0.910 0.912 1.811 (0.664) (0.213) (0.906) (0.309) (0.604) (0.591) Politician positive 1.307 0.867 0.852 1.025 1.506 1.747 (0.440) (0.163) (0.537) (0.358) (0.823) (0.602) Politician negative 2.676** 1.360 1.393 1.290 0.866 1.701 (0.848) (0.263) (0.785) (0.443) (0.534) (0.584) Constant 0.132*** 0.891 0.302 1.384 0.179* 0.457+ (0.053) (0.221) (0.226) (0.654) (0.135) (0.208) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.044 0.044 0.078 0.078 0.089 0.089 Baseline 2 Weeks 4 Weeks VARIABLES Bad Mixed Bad Mixed Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.809 0.883 1.301 1.321 1.170 1.059 (0.579) (0.168) (0.760) (0.447) (0.646) (0.351)  Local negative 2.016* 1.127 1.670 0.910 0.912 1.811 (0.664) (0.213) (0.906) (0.309) (0.604) (0.591) Politician positive 1.307 0.867 0.852 1.025 1.506 1.747 (0.440) (0.163) (0.537) (0.358) (0.823) (0.602) Politician negative 2.676** 1.360 1.393 1.290 0.866 1.701 (0.848) (0.263) (0.785) (0.443) (0.534) (0.584) Constant 0.132*** 0.891 0.302 1.384 0.179* 0.457+ (0.053) (0.221) (0.226) (0.654) (0.135) (0.208) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.044 0.044 0.078 0.078 0.089 0.089 Note: Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. Respondents were asked: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” Response categories were: “good thing,” “bad thing,” and “mixed.” The reference category in the regression model was “good.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. “2 weeks” indicates respondents that took a follow-up survey two weeks after the baseline survey. “4 weeks” indicates respondents that took this follow-up survey four weeks after the original survey. Models also include controls for partisanship, race/ethnicity, education, gender, and age. Table 8. Multinomial Logistic Regressions Predicting Immigration Attitudes (Survey Experiment) Baseline 2 Weeks 4 Weeks VARIABLES Bad Mixed Bad Mixed Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.809 0.883 1.301 1.321 1.170 1.059 (0.579) (0.168) (0.760) (0.447) (0.646) (0.351)  Local negative 2.016* 1.127 1.670 0.910 0.912 1.811 (0.664) (0.213) (0.906) (0.309) (0.604) (0.591) Politician positive 1.307 0.867 0.852 1.025 1.506 1.747 (0.440) (0.163) (0.537) (0.358) (0.823) (0.602) Politician negative 2.676** 1.360 1.393 1.290 0.866 1.701 (0.848) (0.263) (0.785) (0.443) (0.534) (0.584) Constant 0.132*** 0.891 0.302 1.384 0.179* 0.457+ (0.053) (0.221) (0.226) (0.654) (0.135) (0.208) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.044 0.044 0.078 0.078 0.089 0.089 Baseline 2 Weeks 4 Weeks VARIABLES Bad Mixed Bad Mixed Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.809 0.883 1.301 1.321 1.170 1.059 (0.579) (0.168) (0.760) (0.447) (0.646) (0.351)  Local negative 2.016* 1.127 1.670 0.910 0.912 1.811 (0.664) (0.213) (0.906) (0.309) (0.604) (0.591) Politician positive 1.307 0.867 0.852 1.025 1.506 1.747 (0.440) (0.163) (0.537) (0.358) (0.823) (0.602) Politician negative 2.676** 1.360 1.393 1.290 0.866 1.701 (0.848) (0.263) (0.785) (0.443) (0.534) (0.584) Constant 0.132*** 0.891 0.302 1.384 0.179* 0.457+ (0.053) (0.221) (0.226) (0.654) (0.135) (0.208) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.044 0.044 0.078 0.078 0.089 0.089 Note: Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. Respondents were asked: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” Response categories were: “good thing,” “bad thing,” and “mixed.” The reference category in the regression model was “good.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. “2 weeks” indicates respondents that took a follow-up survey two weeks after the baseline survey. “4 weeks” indicates respondents that took this follow-up survey four weeks after the original survey. Models also include controls for partisanship, race/ethnicity, education, gender, and age. Table 8 also shows that the findings for this survey item mirror my previous findings in another key respect: attitudinal effects dissipated in the follow-up surveys. No significant differences were found across treatment conditions for respondents interviewed either two or four weeks after the baseline survey. In the experiment, I also include three more items that seek to capture individual attitudes on three different policy domains: military service, death penalty, and fines against people who pollute the environment. See Appendix A for the wording of these items. Nevertheless, just like I found when analyzing the Gallup survey in the first part of the article, attitudinal effects of immigration statements were confined to immigration items. No spillover effect was detected. See table A4 in the Appendix for full results of reactions to these policies. In summary, the panel survey experiment confirmed that exposure to immigration-related messages can shape attitudes toward immigrants, especially among non-Hispanic whites, conservatives, and lower-educated respondents. Conclusions and Discussion Prior scholars have documented the connection between politicians’ statements and public opinion formation on values, candidates, and public policies. Less is known about politicians’ capacity to affect public attitudes toward groups they demonize, like immigrants, welfare recipients, and criminal offenders. I find that elites’ immigration-related statements have the power to shape public attitudes toward immigrants and identified three structural traits of their effects. First, these effects do not seem to be author-dependent. Second, they are circumscribed to negative statements, as positive ones are not consequential. Third, they are ephemeral. I use two types of methodologies that have unique strengths and complement each other. While the survey analysis is based on a nationally representative sample of the US population and is based on a real-life event, the survey experiment confirms this causal relationship and tests its durability. Politicians’ Statements Are Consequential, but May Not Be Author-Dependent In a previous ethnographic study, I found that the proposal of a high-profile anti-immigrant law by the mayor of Hazleton, PA, appeared to stir anti-immigrant feelings among the native population (Flores 2014). After interviewing more than a hundred Hazleton residents, I theorized that part of this effect could stem from politicians’ public statements. When promoting these laws, restrictionist politicians like the Hazleton mayor or Donald Trump accuse immigrants of abusing social services and other crimes. In that article, I theorized that such statements could harden public opinion toward immigrants (Flores 2014). In this article, I am able to document the independent power of politicians’ negative statements net of any other processes related to the passing of these policies. In light of prior research, it is somewhat surprising that Trump’s explicitly racial statements on immigrants had an attitudinal impact. According to Mendelberg (2001, 191), “the perception exists that for a candidate there is no quicker route to political suicide than being identified as a racist.” So why then were explicitly racial anti-immigrant statements effective? One possibility is that it is more socially acceptable to target racial and ethnic minorities when they are labeled as “immigrants.” Some individuals may believe that the term “immigrants” does not necessarily refer to a specific national-origin group, but it is ethnically neutral (even if Trump specifically referred to “Mexican” immigrants during his campaign). Further, immigrants’ disputed legality may allow political entrepreneurs and their followers to more openly express their racial resentment toward “immigrant” minority groups than toward “native” minorities without being as afraid of violating anti-racist social norms. Yet another possibility is that Trump’s ascendancy to power, based on an explicit rejection of “political correctness,” may usher in a new era in which openly racial language is no longer penalized and instead becomes a legitimate political tool. While I also find that the author of the messages does not seem very consequential, this could have been driven by the study’s limitations. One possibility is that statements by better-known or higher-ranking politicians may be more impactful. In this study, I use a relatively obscure governor to minimize the risk of confounders (i.e., perhaps using a more famous politician would confound the effect of her formal position with that of her personal reputation). However, this is a strict test given that part of the attitudinal power of political elites may be grounded on their fame. In Trump’s case, his fame may counteract the negative effect of his controversial status and produce attitudinal effects. At the same time, even if people accorded no special consideration to politicians’ statements, politicians’ declarations could still be highly consequential since they tend to be broadcasted in the media, as previous research has shown (Page, Shapiro, and Dempsey 1987). Polarity Matters Though prior researchers have mostly theorized about the effect of negative statements about immigration, in this study I go beyond these theoretical expectations and test whether positive messages could also affect public attitudes toward immigrants. I find that positive messages do not influence public attitudes toward immigrants. Why is that the case? One possibility is that these results are driven by social desirability bias. Though, as I suggest above, targeting immigrants may be more permissible than targeting native US minorities, it still may not be fully socially desirable. Nevertheless, upon hearing a public figure like a politician making an anti-immigrant statement, some of these individuals might feel emboldened to express their true, negative views. This would imply that their actual opinion about immigrants does not change. Rather, they become more willing to express their real views as the politician seemingly legitimates them. This “emboldening” theoretical model resonates with my prior work on the effect of punitive immigration laws on immigration attitudes (Flores 2014, 2017). I have found that the passage of these policies increased the willingness of local residents to express anti-immigrant views and become politically active to oppose immigrants. This increased level of anti-immigrant activism created the perception that natives’ views toward immigrants had hardened as a result of the policy even if their underlying immigration attitudes had remained unchanged (Flores 2014, 2017). This would also explain why positive messages about immigration failed to produce attitudinal effects, since pro-immigrant respondents may be less concerned about social desirability and, thus, may be unaffected by politician statements since they do not feel the need to conceal their true feelings. Since US national rhetoric posits that “we are a nation of immigrants” (Alba and Nee 2009), publicly opposing immigrants may be less socially desirable than embracing them. Effects Are Ephemeral This study also illuminates the temporal dimension of this process, as I find that the statements’ effect on public attitudes toward immigrants is short-lived. Social desirability bias may also explain the ephemerality of these effects. If targeting minorities and immigrants is somewhat controversial for some, hearing a political figure openly chastising them may enable some individuals to express their true opinions. However, after a while, this “social permission” may wear off and there would be a return to the status quo, in which such views may still be somewhat controversial. This finding resonates with the literature on political advertisements, which finds that ads by politicians have small, short-term effects on political preferences (Gerber et al. 2011; Hill et al. 2013). This suggests that politicians’ rhetoric shares a fundamental characteristic with political ads. At their very core, both are political statements. The temporality of these effects also resonates with my previous ethnographic study of Hazleton, PA. The proposal of an anti-immigrant policy by the local mayor, Lou Barletta, energized anti-immigrant mobilization by natives (Flores 2014). But these effects declined over time, particularly after Barletta was elected to Congress and local media published fewer stories in which he denounced immigrants (Flores 2014; Longazel 2016). This suggests that to keep the population in a state of excitement, anti-immigrant politicians need to constantly repeat their restrictionist messages to their like-minded bases. This fact has been recognized by Trump himself, who revealed in an interview how he energizes the crowds: “You know, if it gets a little boring, if I see people starting to sort of, maybe thinking about leaving, I can sort of tell the audience, I just say, ‘We will build the wall!’ and they go nuts” (Penzenstadler 2016). Some caveats are in order. Though I show that politicians’ statements about immigration may have short-term attitudinal consequences, the signals I employ use multiple threat dimensions, including crime and tax costs. It is not entirely clear that each threat dimension is equally consequential to immigration attitudes, which future research could investigate. As mentioned earlier, the survey item used by Gallup mentions both “immigrants” and “immigration” in the same question. Nevertheless, these terms may evoke different connotations in people’s minds: immigrant people, immigration as a process, or even immigration policies. This could impact people’s expressed attitudes. For example, given US ideology as a “nation of immigrants” (Alba and Nee 2009), it is possible that individuals have more positive views toward immigrants as a social group than toward immigration policies, which both liberal and conservative politicians routinely criticize (Massey and Sánchez 2010). Future scholars could examine these different connotations and their consequences. Further, the framings used to portray positive messages on immigration may have limited their attitudinal effects. The positive message used, which mentioned criminal and education costs, was designed to closely mirror the negative frames employed by anti-immigrant politicians. Nevertheless, the sole mention of these costs, even if the positive messages questioned them, may have reduced the attitudinal effects of positive messages by reminding survey takers of the likely costs associated with immigration. A different framing could have been used to portray immigration in a positive light, such as human rights (Bloemraad, Silva, and Voss 2016). These caveats aside, however, this paper provides strong evidence that politicians’ immigration-related statements can shape expressed public attitudes toward immigrants, at least in the short term. Footnotes 1 For example, Jeb Bush, long considered the front-runner Republican candidate, did not mention immigration at all during his campaign announcement speech in June 2016 (Bush 2015). 2 A parallel literature in sociology studies collective action frames and framing processes in relation to social movements (see Benford and Snow 2000;,Johnston 1995). 3 My expectation is that most individuals were exposed to Trump’s speech via coverage by newspapers, radio, or TV the day after the event (or at nighttime, which is outside of Gallup’s allotted time to conduct interviews). However, it is possible that some survey respondents may have been exposed to it the same day it took place, either through live TV coverage or by browsing the Internet. I would anticipate that people without regular occupations outside their home were probably more likely to have seen coverage of it the same day it took place. In analyses not shown, I exclude all respondents that were more likely to be home the day of Trump’s announcement, including retired, disabled, and unemployed individuals. In other words, I restrict my sample to individuals who were probably not home the day of the speech and, hence, were more likely to find out about Trump’s speech during the night of June 16 or the morning after, including students, part-time and full-time workers, and unemployed people looking for work. I find that immigration attitudes significantly hardened after Trump’s speech even among these respondents. In addition, in a different set of models, I use an even more conservative approach by excluding respondents interviewed on June 16, the day of the speech, but I continue to find that Republican and non-Hispanic white respondents were significantly more likely to want to “decrease” immigration flows in the post-speech period. 4 News coverage data from other conservative sources like Fox News were not available. Nevertheless, we would expect those sources to pay even more attention to a Republican candidate like Donald Trump. 5 A long-standing sociological literature shows how political actors like groups or politicians use emotions as tools to mobilize individuals (see Bail 2015; Polletta and Jasper 2001). In this case, the angry language used by Trump may have contributed to the viral spread of his message. 6 In models not shown, I found that the relative risk of respondents choosing “same” relative to “decrease” did not significantly change in the post-Trump period. In other words, change took place between “increase” and other categories. 7 Indeed, only one survey respondent was from Wyoming. 8 In these manipulation tests, I found that 98 percent of respondents could correctly remember the identity of the speaker when he was a politician and 82–83 percent could recall it when he was a “local resident.” In addition, 98–99 percent of respondents could correctly recall that the text’s topic was immigration and 96–97 percent correctly recalled the text’s sentiment (whether it was positive or negative). Therefore, I conclude that the different experimental conditions successfully conveyed the treatment cues I intended. 9 Effects do not significantly vary when control variables are not included in the models. 10 Though the results for these interaction tests ran in the same direction as in the survey analysis, only the race interactions were statistically significant at conventional levels, perhaps due to small sample sizes. 11 These respondents include self-identified conservatives and moderates, which accounted for 24.15 percent of the sample. 12 Among respondents re-interviewed four weeks after the baseline survey, 26 percent of individuals assigned to the control condition wanted to “decrease” immigration levels in the United States. Such number was 24 percent and 27 percent for individuals exposed to positive and negative statements by a local resident, respectively. With regard to individuals who read a statement by a politician, 22 percent and 29 percent of those who read positive and negative statements, respectively, also wanted to decrease immigration levels. These differences are not statistically significant at conventional levels. Appendix A Experimental Prompts and Survey Items In this Appendix, we present the prompts for our experimental conditions and then list the survey questions each respondent should answer. Please read the following statement carefully. You will be asked some questions about it afterwards. Experimental conditions (only one was randomly displayed) 1. Positive statements about immigrants from nonpolitician (87 words) During a recent public gathering, Matt Reed, a local resident, said: “The U.S. has become a magnet for many talented people. We have over 10 million undocumented immigrants living in the U.S. but we have the resources to sustain it. It costs us little in terms of health care, education, or incarceration expenses.” Moreover, “because these immigrants are mostly peaceful, our communities are welcoming them. They are bringing new foods. They’re law-abiding. We should support these immigrants since they improve our quality of life,” the resident said. 2. Positive statements about immigrants from politician (88 words) During a recent public gathering, Matt Mead, the governor of Wyoming, said: “The U.S. has become a magnet for many talented people. We have over 10 million undocumented immigrants living in the U.S. but we have the resources to sustain it. It costs us little in terms of health care, education, or incarceration expenses.” Moreover, “because these immigrants are mostly peaceful, our communities are welcoming them. They are bringing new foods. They’re law-abiding. We should support these immigrants since they improve our quality of life,” the governor said. 3. Negative statements about immigrants from nonpolitician (95 words) During a recent public gathering, Matt Reed, a local resident, said: “The U.S. has become a dumping ground for everybody else’s problems. We have over 10 million illegal immigrants living in the U.S. and we do not have the resources to sustain it. It costs us a tremendous amount of money in health care, education, and incarceration.” Moreover, “because of violent acts committed by illegal aliens, our communities are gripped by fear. They are bringing drugs. They’re bringing crime. We cannot stand idly by as these criminals compromise our quality of life,” the resident said. 4. Negative statements about immigrants from politician (96 words) During a recent public gathering, Matt Mead, the governor of Wyoming, said: “The U.S. has become a dumping ground for everybody else’s problems. We have over 10 million illegal immigrants living in the U.S. and we do not have the resources to sustain it. It costs us a tremendous amount of money in health care, education, and incarceration.” Moreover, “because of violent acts committed by illegal aliens, our communities are gripped by fear. They are bringing drugs. They’re bringing crime. We cannot stand idly by as these criminals compromise our quality of life,” the governor said. 5. Control group (89 words) Migration is typically defined as a permanent geographic or spatial change in usual residence from one defined area to another. International migration consists of people changing permanent residence across national boundaries. In reference to the national estimates, net international migration to the United States is the net of 1) migration to the United States and 2) permanent departure from the United States. Temporary international movement of U.S. citizens, including the Armed Forces and their dependents, is not included in calculations of net international migration trends and is treated separately. Thinking now about immigrants—that is, people who come from other countries to live here in the United States, In your view, should immigration be kept at its present level, increased, or decreased? Present level Increased Decreased On the whole, do you think immigration is a good thing or a bad thing for this country today? Good thing Bad thing Mixed Do you favor or oppose the death penalty for persons convicted of murder? Favor Oppose Don’t know Even though they are no longer drafted for military service, young men are still required by law to register for the draft when they become 18 years old. If a young man refuses to register for the draft, do you think he should be punished in any way? Yes No Don’t know Do you favor or oppose to give stiff fines for individuals that pollute the environment? Favor Oppose Don’t know Table A1. Randomization Check (n = 1,317), percent Conditions White Black Hispanic College+ Age 40+ Conservative Neutral (control) 76.0 6.1 7.0 44.4 29.3 18.5 Positive/resident 76.1 5.4 6.5 51.8 22.8 25.7 Negative/resident 75.4 5.0 7.8 52.7 26.6 26.9 Positive/politician 70.7 8.9 10.0 49.6 25.2 22.9 Negative/politician 77.7 6.6 8.2 52.0 27.0 24.1 Conditions White Black Hispanic College+ Age 40+ Conservative Neutral (control) 76.0 6.1 7.0 44.4 29.3 18.5 Positive/resident 76.1 5.4 6.5 51.8 22.8 25.7 Negative/resident 75.4 5.0 7.8 52.7 26.6 26.9 Positive/politician 70.7 8.9 10.0 49.6 25.2 22.9 Negative/politician 77.7 6.6 8.2 52.0 27.0 24.1 Note: “Conservative” refers to self-identified conservatives. “College+” refers to respondents with at least a completed college degree. “White” and “Black” refer to non-Hispanic self-identified whites and blacks, respectively. “Hispanic” refers to self-identified Hispanics who can be of any race. Table A1. Randomization Check (n = 1,317), percent Conditions White Black Hispanic College+ Age 40+ Conservative Neutral (control) 76.0 6.1 7.0 44.4 29.3 18.5 Positive/resident 76.1 5.4 6.5 51.8 22.8 25.7 Negative/resident 75.4 5.0 7.8 52.7 26.6 26.9 Positive/politician 70.7 8.9 10.0 49.6 25.2 22.9 Negative/politician 77.7 6.6 8.2 52.0 27.0 24.1 Conditions White Black Hispanic College+ Age 40+ Conservative Neutral (control) 76.0 6.1 7.0 44.4 29.3 18.5 Positive/resident 76.1 5.4 6.5 51.8 22.8 25.7 Negative/resident 75.4 5.0 7.8 52.7 26.6 26.9 Positive/politician 70.7 8.9 10.0 49.6 25.2 22.9 Negative/politician 77.7 6.6 8.2 52.0 27.0 24.1 Note: “Conservative” refers to self-identified conservatives. “College+” refers to respondents with at least a completed college degree. “White” and “Black” refer to non-Hispanic self-identified whites and blacks, respectively. “Hispanic” refers to self-identified Hispanics who can be of any race. Table A2. Logistic Regression Predicting Responding to Follow-Up Surveys (1) (2) VARIABLES 2 Weeks 4 Weeks Treatment condition (ref = control)  Local positive 0.200 −0.0461 (0.292) (0.273)  Local negative 0.296 −0.048 (0.303) (0.282)  Politician positive −0.053 −0.225 (0.287) (0.271)  Politician negative 0.162 0.000 (0.297) (0.284) Age 0.047*** 0.048*** (0.010) (0.010) Race/ethnicity  Non-Hispanic white −0.508 0.152 (0.352) (0.367)  Non-Hispanic black 0.181 0.256 (0.340) (0.369)  Asian −0.332 0.523 (0.320) (0.328)  Hispanic 0.755 −0.235 (0.662) (0.427)  Other 0.294 0.191 (0.188) (0.176) Conservative −0.194 −0.222 (0.216) (0.207) Community size (ref = large city)  Medium city 0.077 0.044 (0.229) (0.224)  Small town −0.164 −0.064 (0.275) (0.253)  Rural 0.322 0.602 (0.365) (0.371) Constant −0.862+ −0.985* (0.452) (0.422) Observations 638 632 (1) (2) VARIABLES 2 Weeks 4 Weeks Treatment condition (ref = control)  Local positive 0.200 −0.0461 (0.292) (0.273)  Local negative 0.296 −0.048 (0.303) (0.282)  Politician positive −0.053 −0.225 (0.287) (0.271)  Politician negative 0.162 0.000 (0.297) (0.284) Age 0.047*** 0.048*** (0.010) (0.010) Race/ethnicity  Non-Hispanic white −0.508 0.152 (0.352) (0.367)  Non-Hispanic black 0.181 0.256 (0.340) (0.369)  Asian −0.332 0.523 (0.320) (0.328)  Hispanic 0.755 −0.235 (0.662) (0.427)  Other 0.294 0.191 (0.188) (0.176) Conservative −0.194 −0.222 (0.216) (0.207) Community size (ref = large city)  Medium city 0.077 0.044 (0.229) (0.224)  Small town −0.164 −0.064 (0.275) (0.253)  Rural 0.322 0.602 (0.365) (0.371) Constant −0.862+ −0.985* (0.452) (0.422) Observations 638 632 Note: Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 Table A2. Logistic Regression Predicting Responding to Follow-Up Surveys (1) (2) VARIABLES 2 Weeks 4 Weeks Treatment condition (ref = control)  Local positive 0.200 −0.0461 (0.292) (0.273)  Local negative 0.296 −0.048 (0.303) (0.282)  Politician positive −0.053 −0.225 (0.287) (0.271)  Politician negative 0.162 0.000 (0.297) (0.284) Age 0.047*** 0.048*** (0.010) (0.010) Race/ethnicity  Non-Hispanic white −0.508 0.152 (0.352) (0.367)  Non-Hispanic black 0.181 0.256 (0.340) (0.369)  Asian −0.332 0.523 (0.320) (0.328)  Hispanic 0.755 −0.235 (0.662) (0.427)  Other 0.294 0.191 (0.188) (0.176) Conservative −0.194 −0.222 (0.216) (0.207) Community size (ref = large city)  Medium city 0.077 0.044 (0.229) (0.224)  Small town −0.164 −0.064 (0.275) (0.253)  Rural 0.322 0.602 (0.365) (0.371) Constant −0.862+ −0.985* (0.452) (0.422) Observations 638 632 (1) (2) VARIABLES 2 Weeks 4 Weeks Treatment condition (ref = control)  Local positive 0.200 −0.0461 (0.292) (0.273)  Local negative 0.296 −0.048 (0.303) (0.282)  Politician positive −0.053 −0.225 (0.287) (0.271)  Politician negative 0.162 0.000 (0.297) (0.284) Age 0.047*** 0.048*** (0.010) (0.010) Race/ethnicity  Non-Hispanic white −0.508 0.152 (0.352) (0.367)  Non-Hispanic black 0.181 0.256 (0.340) (0.369)  Asian −0.332 0.523 (0.320) (0.328)  Hispanic 0.755 −0.235 (0.662) (0.427)  Other 0.294 0.191 (0.188) (0.176) Conservative −0.194 −0.222 (0.216) (0.207) Community size (ref = large city)  Medium city 0.077 0.044 (0.229) (0.224)  Small town −0.164 −0.064 (0.275) (0.253)  Rural 0.322 0.602 (0.365) (0.371) Constant −0.862+ −0.985* (0.452) (0.422) Observations 638 632 Note: Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 Table A3. Multinomial Regression Models Predicting Immigration Attitudes (Survey Experiment) Baseline 2 weeks 4 weeks VARIABLES Same Decrease Same Decrease Same Decrease Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.229 1.104 1.753 1.254 (0.307) (0.378) (0.548) (0.539) (0.668) (0.570)  Local negative 1.195 2.255** 1.266 0.859 2.743* 1.925 (0.295) (0.631) (0.541) (0.404) (1.164) (0.961)  Politician positive 1.347 1.451 0.978 0.668 2.647* 1.450 (0.317) (0.409) (0.402) (0.317) (1.083) (0.670)  Politician negative 1.098 2.455** 1.265 0.901 3.988** 2.828 (0.271) (0.682) (0.540) (0.428) (1.919) (1.511) Race/ethnicity  Non-Hispanic Black 1.499 1.363 0.987 0.369 3.094 3.598 (0.516) (0.530) (0.525) (0.251) (2.578) (3.215)  Asian 0.984 0.590 2.065 0.978 0.669 0.657 (0.275) (0.214) (0.945) (0.624) (0.330) (0.410)  Hispanic 1.504 0.908 2.323 1.072 0.565 0.694 (0.459) (0.323) (1.373) (0.765) (0.243) (0.351)  Other 0.516 0.439 0.725 0.631 0.869 1.300 (0.194) (0.218) (0.521) (0.528) (0.623) (1.076) Conservative/mod 1.736* 6.679*** 1.584 7.347*** 2.007+ 5.449*** (0.434) (1.688) (0.647) (3.022) (0.827) (2.330) College 0.980 0.573** 0.920 0.621 0.800 0.505* (0.156) (0.103) (0.251) (0.191) (0.224) (0.162) Female 0.865 1.066 0.733 0.856 1.161 0.831 (0.141) (0.194) (0.192) (0.256) (0.316) (0.263) Age 1.020* 1.031*** 1.044** 1.040* 1.017 1.030 (0.008) (0.009) (0.015) (0.017) (0.014) (0.015) Constant 1.337 0.339** 0.539 0.373 0.787 0.345 (0.429) (0.124) (0.334) (0.264) (0.439) (0.220) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.0780 0.0780 0.0924 0.0924 0.0696 0.0696 Baseline 2 weeks 4 weeks VARIABLES Same Decrease Same Decrease Same Decrease Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.229 1.104 1.753 1.254 (0.307) (0.378) (0.548) (0.539) (0.668) (0.570)  Local negative 1.195 2.255** 1.266 0.859 2.743* 1.925 (0.295) (0.631) (0.541) (0.404) (1.164) (0.961)  Politician positive 1.347 1.451 0.978 0.668 2.647* 1.450 (0.317) (0.409) (0.402) (0.317) (1.083) (0.670)  Politician negative 1.098 2.455** 1.265 0.901 3.988** 2.828 (0.271) (0.682) (0.540) (0.428) (1.919) (1.511) Race/ethnicity  Non-Hispanic Black 1.499 1.363 0.987 0.369 3.094 3.598 (0.516) (0.530) (0.525) (0.251) (2.578) (3.215)  Asian 0.984 0.590 2.065 0.978 0.669 0.657 (0.275) (0.214) (0.945) (0.624) (0.330) (0.410)  Hispanic 1.504 0.908 2.323 1.072 0.565 0.694 (0.459) (0.323) (1.373) (0.765) (0.243) (0.351)  Other 0.516 0.439 0.725 0.631 0.869 1.300 (0.194) (0.218) (0.521) (0.528) (0.623) (1.076) Conservative/mod 1.736* 6.679*** 1.584 7.347*** 2.007+ 5.449*** (0.434) (1.688) (0.647) (3.022) (0.827) (2.330) College 0.980 0.573** 0.920 0.621 0.800 0.505* (0.156) (0.103) (0.251) (0.191) (0.224) (0.162) Female 0.865 1.066 0.733 0.856 1.161 0.831 (0.141) (0.194) (0.192) (0.256) (0.316) (0.263) Age 1.020* 1.031*** 1.044** 1.040* 1.017 1.030 (0.008) (0.009) (0.015) (0.017) (0.014) (0.015) Constant 1.337 0.339** 0.539 0.373 0.787 0.345 (0.429) (0.124) (0.334) (0.264) (0.439) (0.220) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.0780 0.0780 0.0924 0.0924 0.0696 0.0696 Note: Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. The question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Response options were: “Present level,” “increased,” and “decreased.” The reference category in the regression model was “increased.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. “2 weeks” indicates respondents that took a follow-up survey two weeks after the baseline survey. “4 weeks” indicates respondents that took this follow-up survey four weeks after the original survey. Table A3. Multinomial Regression Models Predicting Immigration Attitudes (Survey Experiment) Baseline 2 weeks 4 weeks VARIABLES Same Decrease Same Decrease Same Decrease Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.229 1.104 1.753 1.254 (0.307) (0.378) (0.548) (0.539) (0.668) (0.570)  Local negative 1.195 2.255** 1.266 0.859 2.743* 1.925 (0.295) (0.631) (0.541) (0.404) (1.164) (0.961)  Politician positive 1.347 1.451 0.978 0.668 2.647* 1.450 (0.317) (0.409) (0.402) (0.317) (1.083) (0.670)  Politician negative 1.098 2.455** 1.265 0.901 3.988** 2.828 (0.271) (0.682) (0.540) (0.428) (1.919) (1.511) Race/ethnicity  Non-Hispanic Black 1.499 1.363 0.987 0.369 3.094 3.598 (0.516) (0.530) (0.525) (0.251) (2.578) (3.215)  Asian 0.984 0.590 2.065 0.978 0.669 0.657 (0.275) (0.214) (0.945) (0.624) (0.330) (0.410)  Hispanic 1.504 0.908 2.323 1.072 0.565 0.694 (0.459) (0.323) (1.373) (0.765) (0.243) (0.351)  Other 0.516 0.439 0.725 0.631 0.869 1.300 (0.194) (0.218) (0.521) (0.528) (0.623) (1.076) Conservative/mod 1.736* 6.679*** 1.584 7.347*** 2.007+ 5.449*** (0.434) (1.688) (0.647) (3.022) (0.827) (2.330) College 0.980 0.573** 0.920 0.621 0.800 0.505* (0.156) (0.103) (0.251) (0.191) (0.224) (0.162) Female 0.865 1.066 0.733 0.856 1.161 0.831 (0.141) (0.194) (0.192) (0.256) (0.316) (0.263) Age 1.020* 1.031*** 1.044** 1.040* 1.017 1.030 (0.008) (0.009) (0.015) (0.017) (0.014) (0.015) Constant 1.337 0.339** 0.539 0.373 0.787 0.345 (0.429) (0.124) (0.334) (0.264) (0.439) (0.220) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.0780 0.0780 0.0924 0.0924 0.0696 0.0696 Baseline 2 weeks 4 weeks VARIABLES Same Decrease Same Decrease Same Decrease Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.229 1.104 1.753 1.254 (0.307) (0.378) (0.548) (0.539) (0.668) (0.570)  Local negative 1.195 2.255** 1.266 0.859 2.743* 1.925 (0.295) (0.631) (0.541) (0.404) (1.164) (0.961)  Politician positive 1.347 1.451 0.978 0.668 2.647* 1.450 (0.317) (0.409) (0.402) (0.317) (1.083) (0.670)  Politician negative 1.098 2.455** 1.265 0.901 3.988** 2.828 (0.271) (0.682) (0.540) (0.428) (1.919) (1.511) Race/ethnicity  Non-Hispanic Black 1.499 1.363 0.987 0.369 3.094 3.598 (0.516) (0.530) (0.525) (0.251) (2.578) (3.215)  Asian 0.984 0.590 2.065 0.978 0.669 0.657 (0.275) (0.214) (0.945) (0.624) (0.330) (0.410)  Hispanic 1.504 0.908 2.323 1.072 0.565 0.694 (0.459) (0.323) (1.373) (0.765) (0.243) (0.351)  Other 0.516 0.439 0.725 0.631 0.869 1.300 (0.194) (0.218) (0.521) (0.528) (0.623) (1.076) Conservative/mod 1.736* 6.679*** 1.584 7.347*** 2.007+ 5.449*** (0.434) (1.688) (0.647) (3.022) (0.827) (2.330) College 0.980 0.573** 0.920 0.621 0.800 0.505* (0.156) (0.103) (0.251) (0.191) (0.224) (0.162) Female 0.865 1.066 0.733 0.856 1.161 0.831 (0.141) (0.194) (0.192) (0.256) (0.316) (0.263) Age 1.020* 1.031*** 1.044** 1.040* 1.017 1.030 (0.008) (0.009) (0.015) (0.017) (0.014) (0.015) Constant 1.337 0.339** 0.539 0.373 0.787 0.345 (0.429) (0.124) (0.334) (0.264) (0.439) (0.220) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.0780 0.0780 0.0924 0.0924 0.0696 0.0696 Note: Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. The question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Response options were: “Present level,” “increased,” and “decreased.” The reference category in the regression model was “increased.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. “2 weeks” indicates respondents that took a follow-up survey two weeks after the baseline survey. “4 weeks” indicates respondents that took this follow-up survey four weeks after the original survey. Table A4. Logistic Regressions Predicting Responses to Unrelated Policies (Survey Experiment) VARIABLES Death Penalty Pollution Fines Military Service Treatment conditions (ref = neutral)  Local positive 0.309 −0.475 −0.002 (0.174) (0.244) (0.189)  Local negative 0.162 0.138 −0.096 (0.177) (0.275) (0.194)  Politician positive 0.193 −0.180 −0.170 (0.174) (0.256) (0.193)  Politician negative 0.162 0.022 −0.020 (0.177) (0.269) (0.193) Constant −0.178 1.959*** −0.842*** (0.125) (0.189) (0.135) Observations 1,317 1,317 1,317 VARIABLES Death Penalty Pollution Fines Military Service Treatment conditions (ref = neutral)  Local positive 0.309 −0.475 −0.002 (0.174) (0.244) (0.189)  Local negative 0.162 0.138 −0.096 (0.177) (0.275) (0.194)  Politician positive 0.193 −0.180 −0.170 (0.174) (0.256) (0.193)  Politician negative 0.162 0.022 −0.020 (0.177) (0.269) (0.193) Constant −0.178 1.959*** −0.842*** (0.125) (0.189) (0.135) Observations 1,317 1,317 1,317 Note: Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. The question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Response options were: “Present level,” “increased,” and “decreased.” The reference category in the regression model was “increased.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. Table A4. Logistic Regressions Predicting Responses to Unrelated Policies (Survey Experiment) VARIABLES Death Penalty Pollution Fines Military Service Treatment conditions (ref = neutral)  Local positive 0.309 −0.475 −0.002 (0.174) (0.244) (0.189)  Local negative 0.162 0.138 −0.096 (0.177) (0.275) (0.194)  Politician positive 0.193 −0.180 −0.170 (0.174) (0.256) (0.193)  Politician negative 0.162 0.022 −0.020 (0.177) (0.269) (0.193) Constant −0.178 1.959*** −0.842*** (0.125) (0.189) (0.135) Observations 1,317 1,317 1,317 VARIABLES Death Penalty Pollution Fines Military Service Treatment conditions (ref = neutral)  Local positive 0.309 −0.475 −0.002 (0.174) (0.244) (0.189)  Local negative 0.162 0.138 −0.096 (0.177) (0.275) (0.194)  Politician positive 0.193 −0.180 −0.170 (0.174) (0.256) (0.193)  Politician negative 0.162 0.022 −0.020 (0.177) (0.269) (0.193) Constant −0.178 1.959*** −0.842*** (0.125) (0.189) (0.135) Observations 1,317 1,317 1,317 Note: Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. The question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Response options were: “Present level,” “increased,” and “decreased.” The reference category in the regression model was “increased.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. About the Author René D. Flores is a Donald D. Harrington Faculty Fellow at the University of Texas and the Neubauer Family Assistant Professor of Sociology at the University of Chicago. His primary research interests are in the fields of international migration, race and ethnicity, and social stratification. His work has appeared in the American Journal of Sociology, American Sociological Review, Social Forces, and Social Problems. References Alba , Richard , and Victor Nee . 2009 . Remaking the American Mainstream: Assimilation and Contemporary Immigration . Cambridge, MA : Harvard University Press . Arceneaux , Kevin . 2008 . “ Can Partisan Cues Diminish Democratic Accountability? ” Political Behavior 30 ( 2 ): 139 – 60 . Google Scholar CrossRef Search ADS Bandler , Aaron . 2016 . “Here’s a Full Timeline of Donald Trump’s Immigration Positions.” http://www.dailywire.com/news/8644/heres-full-timeline-donald-trumps-immigration-aaron-bandler, accessed June 6, 2017. Bail , Christopher A. 2015 . Terrified: How Anti-Muslim Fringe Organizations Became Mainstream . Princeton, NJ : Princeton University Press . Google Scholar CrossRef Search ADS Barabas , Jason , and Jennifer Jerit . 2010 . “ Are Survey Experiments Externally Valid? ” American Political Science Review 104 : 226 – 42 . Google Scholar CrossRef Search ADS Beckett , Katherine . 1997 . Making Crime Pay: Law and Order in Contemporary American Politics . New York : Oxford University Press . Benford , Robert D. , and David A. Snow . 2000 . “ Framing Processes and Social Movements: An Overview and Assessment .” Annual Review of Sociology 26 ( 1 ): 611 – 39 . Google Scholar CrossRef Search ADS Blinder , S. , R. Ford , and E. Ivarsflaten . 2013 . “ The Better Angels of Our Nature: How the Antiprejudice Norm Affects Policy and Party Preferences in Great Britain and Germany .” American Journal of Political Science 57 ( 4 ): 841 – 57 . Bloemraad , Irene , Fabiana Silva , and Kim Voss . 2016 . “ Rights, Economics, or Family? Frame Resonance, Political Ideology, and the Immigrant Rights Movement .” Social Forces 94 ( 4 ): 1647 – 74 . Google Scholar CrossRef Search ADS Bonilla-Silva , Eduardo . 2003 . “ ‘New Racism,’ Color-Blind Racism, and the Future of Whiteness in America .” White Out: The Continuing Significance of Racism , 271 – 84 . Brown , Hana E. 2013 . “ Race, Legality, and the Social Policy Consequences of Anti-Immigration Mobilization .” American Sociological Review 78 ( 2 ): 290 – 314 . Google Scholar CrossRef Search ADS Burke , Michael . 2016 . “Activity among White Supremacists Continues to Surge.” USA Today, June 16. Bush , Jeb . 2015 . Full text of Jeb Bush’s presidential announcement. http://www.politico.com/story/2015/06/jeb-bush-2016-announcement-full-text-119023, accessed December 17, 2016. Calavita , Kitty . 1996 . “ The New Politics of Immigration: Balanced-Budget Conservatism and the Symbolism of Proposition 187 .” Social Problems 43 : 284 – 305 . Google Scholar CrossRef Search ADS Carroll , Rory . 2016 . “‘You Were Born in a Taco Bell’: Trump’s Rhetoric Fuels School Bullies Across US.” The Guardian, June 9. Chavez , Leo R. 2001 . Covering Immigration: Popular Images and the Politics of the Nation . Berkeley : University of California Press . ——— . 2008 . The Latino Threat: Constructing Immigrants, Citizens, and the Nation . Redwood City, CA : Stanford University Press . Citrin , J. , B. Reingold , and D. P. Green. 1990 . “ American Identity and the Politics of Ethnic Change .” Journal of Politics 57 ( 4 ): 1124 – 54 . Google Scholar CrossRef Search ADS Confessore , Nicholas , and Karen Yourish . 2016 . “$2 Billion Worth of Free Media for Donald Trump.” New York Times, March 15. Druckman , James N. 2001 . “ On the Limits of Framing Effects: Who Can Frame? ” Journal of Politics 63 ( 4 ): 1041 – 66 . Google Scholar CrossRef Search ADS Druckman , James N. , and Kjersten R. Nelson . 2003 . “ Framing and Deliberation: How Citizens’ Conversations Limit Elite Influence .” American Journal of Political Science 47 ( 4 ): 729 – 45 . Google Scholar CrossRef Search ADS Druckman , J. N. , E. Peterson , and R. Slothuus . 2013 . “ How Elite Partisan Polarization Affects Public Opinion Formation .” American Political Science Review 107 ( 1 ): 57 – 79 . Google Scholar CrossRef Search ADS Easton , D. , and J. Dennis . 1969 . Children in the Political System . New York : McGraw-Hill . Edelman , Murray . 1977 . Political Language: Words That Succeed and Policies That Fail . New York : Academic Press . Ehrenfreund , Max . 2016 . “What’s Incredible About Republicans’ Views on Immigration Is How Much They’ve Changed.” Washington Post, February 3. Flores , René D. 2014 . “ In the Eye of the Storm: How Did Hazleton’s Restrictive Immigration Ordinance Affect Local Interethnic Relations? ” American Behavioral Scientist 58 ( 13 ): 1743 – 63 . Google Scholar CrossRef Search ADS ——— . 2015 . “ Taking the Law into Their Own Hands: Do Local Anti-Immigrant Ordinances Increase Gun Sales? ” Social Problems 62 ( 3 ): 363 – 90 . Google Scholar CrossRef Search ADS ——— . 2017 . “ Do Anti-Immigrant Laws Shape Public Sentiment?: A Study of Arizona’s SB 1070 Using Twitter Data .” American Journal of Sociology 123 ( 2 ): 333 – 84 . Google Scholar CrossRef Search ADS Fox , Lauren . 2015 . “Boston Brothers Held on High Bail in Alleged Hate Crime.” Boston Globe, November 26. Franke-Ruta , Garance . 2013 . “What You Need to Read in the RNC Election-Autopsy Report.” The Atlantic, March 18. Gelman , Andrew , and Jennifer Hill . 2006 . Data Analysis Using Regression and Multilevel/Hierarchical Models . Cambridge : Cambridge University Press . Google Scholar CrossRef Search ADS Gerber , Alan S. , James G. Gimpel , Donald P. Green , and Daron R. Shaw . 2011 . “ How Large and Long-Lasting Are the Persuasive Effects of Televised Campaign Ads? Results from a Randomized Field Experiment .” American Political Science Review 105 ( 1 ): 135 – 50 . Google Scholar CrossRef Search ADS Goren , Paul , Christopher M. Federico , and Miki Caul Kittilson . 2009 . “ Source Cues, Partisan Identities, and Political Value Expression .” American Journal of Political Science 53 ( 4 ): 805 – 20 . Google Scholar CrossRef Search ADS Haberman , Maggie . 2016 . “Donald Trump’s Immigration Message May Resound in New Hampshire.” New York Times, February 5. Haberman , Maggie , and Alexander Burns . 2016 . “Donald Trump’s Presidential Run Began in an Effort to Gain Stature.” New York Times, March 12. Hainmueller , Jens , and Daniel J. Hopkins . 2014 . “ Public Attitudes toward Immigration .” Annual Review of Political Science 17 : 225 – 49 . Google Scholar CrossRef Search ADS Hill , Seth J. , James Lo , Lynn Vavreck , and John Zaller . 2013 . “ How Quickly We Forget: The Duration of Persuasion Effects from Mass Communication .” Political Communication 30 ( 4 ): 521 – 47 . Google Scholar CrossRef Search ADS Hopkins , D. J. 2010 . “ Politicized Places: Explaining Where and When Immigrants Provoke Local Opposition .” American Political Science Review 104 ( 1 ): 40 – 60 . Google Scholar CrossRef Search ADS Hopkins , Daniel J. , Van C. Tran , and Abigail F. Williamson . 2014 . “ See No Spanish: Language, Local Context, and Attitudes toward Immigration .” Politics, Groups, and Identities 2 ( 1 ): 35 – 51 . Google Scholar CrossRef Search ADS Hyman , H. H. 1959 . Political Socialization . New York : Free Press . Johnston , Hank . 1995 . “ A Methodology for Frame Analysis: From Discourse to Cognitive Schemata .” Social Movements and Culture 4 : 2l7 – 46 . Joslyn , Mark R. , and Donald P. Haider-Markel . 2006 . “ Should We Really ‘Kill’ the Messenger? Framing Physician-Assisted Suicide and the Role of Messengers .” Political Communication 23 ( 1 ): 85 – 103 . Google Scholar CrossRef Search ADS Kam , Cindy D. 2005 . “ Who Toes the Party Line? Cues, Values, and Individual Differences .” Political Behavior 27 ( 2 ): 163 – 82 . Google Scholar CrossRef Search ADS Krieg , Gregory . 2016 . “14 of Trump’s Most Outrageous ‘Birther’ Claims—Half from after 2011.” http://www.cnn.com/2016/09/09/politics/donald-trump-birther/index.html, accessed June 6, 2017. Kuklinski , James H. , and Norman L. Hurley . 1994 . “ On Hearing and Interpreting Political Messages: A Cautionary Tale of Citizen Cue-Taking .” Journal of Politics 56 ( 3 ): 729 – 51 . Google Scholar CrossRef Search ADS Legewie , J. 2013 . “ Terrorist Events and Attitudes toward Immigrants: A Natural Experiment .” American Journal of Sociology 118 ( 5 ): 1199 – 1245 . Google Scholar CrossRef Search ADS Longazel , Jamie . 2016 . Undocumented Fears: Immigration and the Politics of Divide and Conquer in Hazleton, Pennsylvania . Philadelphia : Temple University Press . Marrow , Helen B. 2011 . New Destination Dreaming: Immigration, Race, and Legal Status in the Rural American South . Redwood City, CA : Stanford University Press . Massey , Douglas S. , and Karen A. Pren . 2012 . “ Origins of the New Latino Underclass .” Race and Social Problems 4 ( 1 ): 5 – 17 . Google Scholar CrossRef Search ADS PubMed Massey , Douglas , and Magaly Sánchez . 2010 . Brokered Boundaries: Creating Immigrant Identity in Anti-Immigrant Times . New York : Russell Sage Foundation . Mendelberg , Tali . 2001 . The Race Card: Campaign Strategy, Implicit Messages, and the Norm of Equality . Princeton, NJ : Princeton University Press . Google Scholar CrossRef Search ADS Nelson , Thomas E. , and Donald R. Kinder . 1996 . “ Issue Frames and Group-Centrism in American Public Opinion .” Journal of Politics 58( 4 ): 1055 – 78 . Google Scholar CrossRef Search ADS Nicholson , Stephen P. 2011 . “ Dominating Cues and the Limits of Elite Influence .” Journal of Politics 73 ( 4 ): 1165 – 77 . Google Scholar CrossRef Search ADS ——— . 2012 . “ Polarizing Cues .” American Journal of Political Science 56 ( 1 ): 52 – 66 . Google Scholar CrossRef Search ADS PubMed Page , Benjamin I. , Robert Y. Shapiro , and Glenn R. Dempsey . 1987 . “ What Moves Public Opinion? ” American Political Science Review 81 ( 1 ): 23 – 43 . Google Scholar CrossRef Search ADS Penzenstadler , Nick . 2016 . “Trump: When Audiences Get Bored I Use ‘The Wall’.” USA TODAY. https://www.usatoday.com/story/news/politics/onpolitics/2016/01/30/trump-when-audiences-get-bored-use-wall/79573388/. Polletta , Francesca , and James M. Jasper . 2001 . “ Collective Identity and Social Movements .” Annual Review of Sociology 27 ( 1 ): 283 – 305 . Google Scholar CrossRef Search ADS Republican National Committee . 2013 . Growth and Opportunity Project. July 16. Rove , Karl . 2013 . “More White Votes Alone Won’t Save the GOP: To Win the Presidency in 2016, the Party Needs to Do Better with Hispanics.” Wall Street Journal, June 26. Rubin , Jennier . 2013 . “GOP Autopsy Report Goes Bold.” Washington Post, March 18. Sanneh , Kelefa . 2015 . “A Serious Immigration Debate, Thanks to Donald Trump.” New Yorker, August 19. Santa Ana , Otto . 2002 . Brown Tide Rising: Metaphoric Representations of Latinos in Contemporary American Public Discourse . Austin : University of Texas Press . Schachter , Ariela . 2015 . “A Change of Heart or Change of Address? The Geographic Sorting of Whites’ Attitudes towards Immigration.” Unpublished manuscript. Sears , D. O. 1993 . “Symbolic Politics: A Socio-Psychological Theory.” In Explorations in Political Psychology , edited by S. Iyengar and W. J. McGuire , 113 – 49 . Durham, NC : Duke University Press . Slothuus , Rune . 2010 . “ When Can Political Parties Lead Public Opinion? Evidence from a Natural Experiment .” Political Communication 27 ( 2 ): 158 – 77 . Google Scholar CrossRef Search ADS Soss , J. , and S. F. Schram . 2007 . “ A Public Transformed? Welfare Reform as Policy Feedback .” American Political Science Review 101 ( 1 ): 111 – 27 . Google Scholar CrossRef Search ADS Sunstein , Cass . 1996 . “ On the Expressive Function of Law .” University of Pennsylvania Law Review 144 : 2021 – 53 . Google Scholar CrossRef Search ADS Terkildsen , Nayda , and Frauke Schnell . 1997 . “ How Media Frames Move Public Opinion: An Analysis of the Women’s Movement .” Political Research Quarterly 50 ( 4 ): 879 – 900 . Google Scholar CrossRef Search ADS Trump , Donald . 2015 . “Presidential Announcement Speech.” June 16. http://time.com/3923128/donald-trump-announcement-speech/. Vasquez , Tina . 2015 . “I’ve Experienced a New Level of Racism Since Donald Trump Went After Latinos.” The Guardian, September 9. Weinberg , Jill D. , Jeremy Freese , and David McElhattan . 2014 . “ Comparing Data Characteristics and Results of an Online Factorial Survey between a Population-Based and a Crowdsource-Recruited Sample .” Sociological Science 1 : 292 – 310 . Google Scholar CrossRef Search ADS Author notes I thank Hana Brown and Emilce Santana for generously providing careful review and valuable comments. I also thank Amanda Mireles for her help securing data access. Jenefer Jedele and Peter Chu provided research assistance. All errors are uniquely my own. © The Author(s) 2018. Published by Oxford University Press on behalf of the University of North Carolina at Chapel Hill. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Social Forces Oxford University Press

Can Elites Shape Public Attitudes Toward Immigrants?: Evidence from the 2016 US Presidential Election

Social Forces , Volume Advance Article (4) – Apr 27, 2018

Loading next page...
 
/lp/ou_press/can-elites-shape-public-attitudes-toward-immigrants-evidence-from-the-K6fP4XHkUS
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of the University of North Carolina at Chapel Hill. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
ISSN
0037-7732
eISSN
1534-7605
D.O.I.
10.1093/sf/soy001
Publisher site
See Article on Publisher Site

Abstract

Abstract It is well known that political elites can shape public attitudes toward policies and values. Less is known, however, about whether elites can also influence public perceptions of social groups they praise or denounce. I test this by analyzing the attitudinal effects of Donald Trump’s 2016 presidential campaign announcement speech, in which he referred to Mexican immigrants as “rapists” and “criminals.” First, to provide causal estimates, I analyze survey data using a counterfactual approach. I find evidence that Trump’s statements negatively affected public opinion toward immigrants particularly among groups with restrictionist tendencies. Second, using a panel survey experiment, I corroborate this causal relationship but find that these effects are short-lived. This explains why restrictionist politicians like Trump constantly prod natives to keep their messages’ effects from dissipating. I also find that only negative messages are consequential and find no evidence that elite statements are more impactful than those from non-elites, suggesting that the power of elite rhetoric lies primarily in its capacity to reach the masses via the news media. Introduction It is well known that elites can shape public attitudes toward policies, politicians, and belief statements (Mendelberg 2001). Less is known, however, about whether political elites can also influence public perceptions of the social groups they praise or denounce. Scholars have argued that politicians’ use of symbolic language blaming vulnerable groups, such as racial minorities, immigrants, and poor families, for society’s problems may not only encourage popular support for exclusionary policies but also influence public views of these groups themselves (Beckett 1997; Calavita 1996; Chavez 2008; Citrin, Reingold, and Green 1990; Edelman 1977; Santa Ana 2002). Though this hypothesis has been applied to the study of multiple policy domains, including welfare, crime/policing, immigration, and family (Beckett 1997; Chavez 2001; Flores 2015; Soss and Schram 2007), there is relatively little direct evidence that elites’ statements do in fact shape public attitudes toward targeted groups. In this paper, I test this widespread assumption by assessing the effect of anti-immigrant rhetoric by political elites on public attitudes toward immigrants as a case study of this long-held hypothesis. Assessing the effect of politicians’ rhetoric on immigration is not an easy task. Some of this complexity stems from the fact that immigration goes through cycles of low and high politicization (Massey and Pren 2012). In periods of high politicization, it becomes difficult to assess the power of a specific politician’s statements due to the large number of public figures opining about immigration simultaneously. Further, politicians’ statements may be endogenous to public mood about migration. Nevertheless, Donald Trump’s 2016 presidential campaign may provide an empirical opportunity to do so. On June 16, 2015, Trump announced his presidential candidacy in a speech in which he explicitly referred to Mexican immigrants as “rapists” and “criminals.” This speech received substantial media coverage (Burke 2016; Confessore and Yourish 2016). Since, up until that point, no other presidential candidate during the 2016 election cycle had made immigration a central campaign topic1 (Sanneh 2015), Trump’s speech provides an ideal empirical opportunity to assess whether politicians’ statements about immigrants do in fact shape immigration attitudes. I use two distinct methodologies that have unique strengths to address this question. First, I use a Gallup survey that was on the field during Trump’s speech and I compare the immigration attitudes of respondents interviewed right before the speech with those interviewed right after. Because the survey’s day of application was random, it allows me to approximate a natural experiment by treating respondents as if they were randomly assigned to pre- and post-treatment groups. I find evidence that Trump’s negative statements negatively affected public opinion toward immigrants particularly among Republicans and individuals without college degrees. While this survey analysis addresses concerns about external validity and realism by design, since the sample used is nationally representative and it examines a real life event, it cannot disentangle whether the effect was driven by Trump’s original speech or by subsequent media coverage or whether it led to permanent or temporary attitudinal changes. To address these questions, I conduct a national panel survey experiment in which I follow respondents that were exposed to both pro- and anti-immigrant messages over several weeks to test the duration of these messages’ attitudinal effects. Not only does the panel survey experiment corroborate the causal link between politicians’ statements and immigration attitudes, but it also uncovers three structural traits. First, these effects are ephemeral. They dissipate within days. Second, these effects are circumscribed to negative statements, as positive ones do not seem consequential. Third, they are not author-dependent. Statements by politicians were not more impactful than statements by local residents. This suggests that the power of elite rhetoric primarily lies in its capacity to reach the masses via the news media. Prior scholars have argued that elite statements that are explicitly racial in nature, like Trump’s speech on immigration, may fail to have attitudinal effects given widely held social norms against the open expression of racial prejudice (Mendelberg 2001). Nevertheless, my findings imply that immigration statements may be an exception. Immigrants’ contested legality may allow critics to explicitly target them without seemingly violating anti-racist social norms. Political Elites and Public Opinion There is ample evidence that political elites can influence public opinion. Researchers have found that elites can affect public attitudes toward laws (Druckman 2001; Mendelberg 2001; Nicholson 2011, 2012); values and statements (Goren, Federico, and Kittilson 2009; Kuklinski and Hurley 1994); political candidates (Arceneaux 2008); and even medical practices (Joslyn and Haider-Markel 2006). We know less, however, about elites’ capacity to influence public opinion about social groups they extol or condemn. Public opinion toward policies may be different from attitudes toward social groups. For example, both conservative and liberal elites typically describe “immigration laws” as “broken” (Massey and Sánchez 2010). In contrast, since national ideology defines the United States as a “nation of immigrants,” mainstream elites tend to portray “immigrants” in a positive light and highlight their contributions (Alba and Nee 2009). The interdisciplinary “symbolic politics” perspective suggests that elite statements may shape public views of targeted groups. From this perspective, individuals acquire affective predispositions such as ethnocentrism, racial attitudes, and altruism through socialization early in life. These predispositions guide their attitudes toward social and political issues (Easton and Dennis 1969; Hyman 1959; Sears 1993). Politicians employ symbols, images, and metaphors to evoke these predispositions and mobilize public sentiment about specific issues (Beckett 1997). When promoting divisive public policies, politicians use symbolic language that implicitly identifies social groups, such as racial minorities, immigrants, and poor families, as the source of social ailments (Beckett 1997; Calavita 1996; Edelman 1977). When these symbolic appeals connect with people’s emotional predispositions, the general public will often rally around punitive policies that target specific subgroups (Sears 1993). In addition to encouraging public support for exclusionary policies, scholars have claimed that such symbolic political discourse may also shape public views of the targeted groups (Calavita 1996; Chavez 2001, 2008; Santa Ana 2002). However, previous studies have not firmly established whether restrictionist politicians are causing or merely echoing public opinion. It is certainly plausible that politicians’ statements could have an independent effect on public opinion toward minorities, but it is equally possible that they merely echo the views of the general population by adopting anti-minority stances. A related literature has examined the impact of “frames” used by elites on public attitudes in a more empirical fashion2 (Nelson and Kinder 1996). These studies define framing as “alternative conceptualizations of an issue or event” and have found that, under certain conditions, frames could shape public opinion (Druckman and Nelson 2003). For example, referring to anti-poverty programs as “welfare” reduces the public’s approval of the programs, but framing those same programs as “helping the poor” increases popular support, since the latter label taps into an altruistic predisposition among the public. Nevertheless, though frames commonly enter political discourse via political actors like politicians or political parties, most framing studies have relied on newspaper vignettes or unattributed statements to expose individuals to different frames (Druckman, Peterson, and Slothuus 2013). Some studies have assessed the impact of political parties, most often via party endorsements, on public attitudes (Kam 2005; Slothuus 2010). This literature suggests that political parties may be most effective at shaping their own members’ attitudes. However, it is not entirely clear that this finding extends also to individual politicians who, as individuals, may or may not be seen as true representatives of their entire parties. While scholars theorize that the statements of public figures like politicians or judges may be more consequential than the words of common people (Edelman 1977; Sunstein 1996), evidence about this is mixed (Joslyn and Haider-Markel 2006; Kuklinski and Hurley 1994). This raises questions about whether political statements are always consequential regardless of their author or source. The characteristics of the message may also matter. Mendelberg (2001) examines the effect of political statements on individuals’ policy positions on race, welfare, poverty, spending, and defense. She finds that political messages that are only implicitly racial are more effective in shaping public opinion than those in which race is explicitly mentioned. She argues that elite statements that make “direct verbal references to race” will backfire since they violate the norm of equality. Other authors have similarly claimed that in the present time period, there is a strong social norm against the open expression of racial or ethnic prejudice (Blinder, Ford, and Ivarsflaten 2013; Bonilla-Silva 2003). Therefore, contemporary US politicians will instead use covert or coded language when blaming minorities for societal problems in an attempt to mobilize the majority population. According to this perspective, if an elite statement is obviously racial, as in Donald Trump’s statement, we should expect that this social norm will become activated and neutralize the statements’ effects since most people do not want to be seen as prejudiced (Mendelberg 2001). Public Opinion Toward Immigration Understanding public attitudes toward immigrants is a key area of social science research (Bloemraad, Silva, and Voss 2016; Marrow 2011; Schachter 2015). This vast literature has been generally oriented toward two camps. A materialist camp, with deep roots in economics, argues that the way natives respond to immigrants is deeply colored by their own individual material self-interest (Hainmueller and Hopkins 2014). A second approach, pursued by sociologists and political scientists, finds that public opinion toward immigration is rooted in its perceived cultural and economic impacts on the nation as a whole (Citrin, Reingold, and Green 1990; Hopkins, Tran, and Williamson 2014). This study is not designed to adjudicate between these two competing literatures. Rather, it aims to provide causal evidence for one of the key mechanisms through which cultural or economic concerns presumably become activated in the minds of the general people: elites’ public statements. A significant number of studies assume, but do not directly test, that politicians have the power to shape public attitudes by constructing narratives through which immigrants are viewed (Hopkins 2010; Massey and Sánchez 2010; Santa Ana 2002). Politicians are assumed to be “external, politicizing agents” that have the power to “politicize people’s day-to-day experiences” and even make “people’s views turn anti-immigrant” (Hopkins 2010). Though there is ample evidence that politicians can elect to portray immigrants in positive or negative ways (Brown 2013; Massey and Pren 2012; Santa Ana 2002), there is less direct evidence that their statements do in fact influence public opinion, which I test in this paper. In a previous paper, I adopted a counterfactual approach to examine the effect of SB 1070, a high-profile anti-immigrant law passed by Arizona in 2010, on both public attitudes and behaviors toward immigrants using Twitter data from that state. I found that SB 1070 had a negative impact on the sentiment of the average tweet regarding immigrants, Mexicans, and Hispanics, but not on tweets about Asians or Blacks (Flores 2017). Though that paper demonstrated that restrictionist policies could set into motion attitudinal dynamics, I could not determine whether the effects were driven by the tough language used by anti-immigrant politicians in Arizona or by the passage of the policy itself, which may have legitimated extreme views. In this paper, I address this gap by employing a causal inference approach to examine whether politicians’ statements by themselves can have an independent effect on attitudes toward immigrants. Immigration and the 2016 Presidential Election On June 16, 2015, Donald Trump held a campaign rally in New York City to announce he was running for president of the United States of America. Though Trump was primarily known as a businessman and television personality, he had been personally involved in politics by endorsing candidates, donating money, and publicly promoting his policy preferences for decades (Haberman and Burns 2016). He had also run for office multiple times starting in 2000, when he first campaigned for the US presidency and won two Reform Party primaries. Trump had a history of making racially charged public statements, such as his accusation that President Barack Obama was not born in the United States (Krieg 2016). However, when it came to immigration, his ideological position had constantly shifted. In August 2013, he had actually declared his support for amnesty for undocumented immigrants (Bandler 2016). Up until that point, no other presidential candidates in the 2016 campaign had made immigration a central issue (Sanneh 2015). This was partly the result of a concerted effort by the Republican leadership to avoid taking overly punitive stances on immigration (Haberman 2016; Republican National Committee 2013). Such efforts were conceived by Republican leaders after Mitt Romney lost to Barack Obama and only obtained 27 percent of the Hispanic vote in the 2012 presidential election. These leaders believed that Romney’s hard-line position on immigration and especially his call for the “self-deportation” of millions of undocumented immigrants had alienated Hispanic voters (Franke-Ruta 2013; Rove 2013; Rubin 2013). Nevertheless, Trump broke with the party line by bringing immigration into the forefront in his announcement speech. Standing in front of journalists from all over the country, he said: “When Mexico sends its people, they’re not sending their best. They’re not sending you. They’re not sending you. They’re sending people that have lots of problems, and they’re bringing those problems with us. They’re bringing drugs. They’re bringing crime. They’re rapists. And some, I assume, are good people. But I speak to border guards and they tell us what we’re getting. And it only makes common sense. It only makes common sense. They’re sending us not the right people. It’s coming from more than Mexico. It’s coming from all over South and Latin America, and it’s coming probably—probably—from the Middle East. But we don’t know. Because we have no protection and we have no competence, we don’t know what’s happening. And it’s got to stop and it’s got to stop fast” (Trump 2015). Trump’s speech, which was widely covered by the media, proved influential. Eventually, all the other Republican candidates, with the exception of Jeb Bush, took a hard-line approach to immigration (Ehrenfreund 2016). Since no other prominent politician was talking about immigration prior to his speech, Trump’s announcement provides an ideal opportunity for social scientists to study the consequences of politicians’ statements on immigration attitudes. Since he was the only prominent politician publicly taking a hard-line position on immigration, his statements could be considered an independent shock on public opinion toward immigration. If multiple politicians were making similar statements concomitantly, it would be hard to determine whether any of them was single-handedly affecting public sentiment. Journalists and other observers have claimed that Trump’s speech was deeply consequential. Critics have claimed that it energized white supremacists and hardened public attitudes toward immigrants (Burke 2016; Carroll 2016; Haberman 2016). Immigrants and even US-born Hispanics have reported increased public harassment following Trump’s speech, which includes strangers asking them if they speak English or inquiring about their citizenship status (Vasquez 2015). Further, actual violence has been directly linked to this speech. In August 2015, two white men beat with a pole and urinated on a 58-year-old homeless Mexican man in Boston. The men cited Trump as the reason why they did it. “Trump was right” about “deporting all these illegals,” they said (Fox 2015; Vasquez 2015). Nevertheless, beyond this anecdotal evidence, no systematic study has been conducted to examine the social consequences of Trump’s pronouncements. After all, these incidents could be isolated cases. In addition, there are theoretical reasons why this speech may not have been socially consequential. As mentioned earlier, scholars have suggested that when elite statements are openly racial, like in Trump’s case, attitudinal effects may not occur. This is so because such overt racial language may activate the social norm against the open expression of racial prejudice among individuals. In this paper, I employ two distinct methodologies and use Trump’s announcement speech as a quasi-natural experiment to systematically assess whether elites’ pronouncements on immigration can shape public attitudes toward immigrants. In addition, to address aforementioned literature gaps, I test whether these effects are temporary or durable and whether the identity of the speaker matters. Data and Methods As mentioned earlier, Donald Trump gave his presidential announcement speech on June 16, 2015. By happenstance, Gallup was conducting a nationwide survey on minority rights and relations during that week. The survey included the most commonly used question to gauge attitudes toward immigrants: “Thinking now about immigrants—that is, people who come from other countries to live here in the United States. In your view, should immigration be kept at its present level, increased, or decreased?” The availability of this data and, most importantly, its collection timing, allow me to conduct a counterfactual analysis similar to a regression discontinuity design in which the date of the interview was randomly assigned to respondents (Legewie 2013). By comparing the responses of survey takers that were interviewed right before Trump’s speech with those that Gallup reached slightly after, I can estimate the effect of Donald Trump’s announcement on public attitudes toward immigrants. However, for such comparison to provide quasi-causal estimates, I must show that there was no selection into treatment. In other words, aside from the timing of the interview, both groups of respondents did not differ in any meaningful way that could have affected their immigration views. Gallup began polling on June 15 and ended on July 10. Three hundred fifty-eight respondents were interviewed on June 15 and 16. These respondents form the control group in my study. Trump’s speech took place on June 16, but most media outlets covered his speech on June 17. Indeed, figure 1 shows that mentions of Donald Trump in the seven largest US newspapers increased substantially only on June 17. Therefore, I consider people interviewed on June 15 and 16 to be a good indicator of the prevailing mood about immigration in the United States before Trump’s speech.3 A total of 769 individuals were interviewed in the two following days (June 17–18). This group constitutes the treated group since they were likely exposed to Trump’s rhetoric. For this design to be valid, it needs to meet several conditions. First, there has to be a discontinuity in the treatment. In this case, I need to show that after Trump’s speech, popular media picked up his speech and, hence, individuals in the treated group were more likely to be exposed to more negative portrayals of immigrants. Figure 1. View largeDownload slide Newspaper mentions of Donald Trump (June 2015) Note: The blue line shows the total number of articles mentioning Donald Trump published by the seven largest newspapers in the United States: the New York Times, Los Angeles Times, Washington Post, Wall Street Journal, Boston Globe, USA Today, and Orange County Register. Vertical line indicates the day when Donald Trump launched his presidential campaign (June 16, 2015). The horizontal scale, ranging from 9 to 23, indicates the day of the month. Figure 1. View largeDownload slide Newspaper mentions of Donald Trump (June 2015) Note: The blue line shows the total number of articles mentioning Donald Trump published by the seven largest newspapers in the United States: the New York Times, Los Angeles Times, Washington Post, Wall Street Journal, Boston Globe, USA Today, and Orange County Register. Vertical line indicates the day when Donald Trump launched his presidential campaign (June 16, 2015). The horizontal scale, ranging from 9 to 23, indicates the day of the month. According to the literature, for a politician’s statement to be socially consequential, it has to receive media coverage. To test that, I examine media coverage from the seven largest newspapers in the United States: the New York Times, Los Angeles Times, Washington Post, Wall Street Journal, Boston Globe, USA Today, and Orange County Register. These newspapers cover a wide range of the mainstream ideological spectrum. In addition, given that some individuals may not read newspapers but instead get their news from radio, TV, or Internet sources, I also analyze transcripts from all radio shows produced by the non-partisan National Public Radio (NPR) as well as transcripts from TV news and online articles produced by the Cable News Corporation (CNN).4 Figure 2 shows the number of news items related to immigration produced by each of these media outlets. The vertical line indicates when Donald Trump launched his presidential campaign. O, in this case, represents the week when Trump gave his speech. The horizontal scale, ranging from −4 to +4, indicates the number of weeks since Trump’s announcement. Figure 2 indicates that the absolute number of news items related to immigration increased in six out of nine news outlets after Trump’s announcement. This increase of media coverage is statistically significant at the 99 percent level. The emotional language used by Trump in referring to immigrants may have contributed to the rapid spread of his message (Bail 2015).5 Figure 2. View largeDownload slide Media coverage of immigration (May 19–July 13, 2015)Note: Lines show the total number of articles related to immigration published by each of the seven largest newspapers in the United States: the New York Times, Los Angeles Times, Washington Post, Wall Street Journal, Boston Globe, USA Today, and Orange County Register. It also includes news coverage data from all radio shows produced by National Public Radio (NPR) as well as transcripts from TV news and online articles produced by the Cable News Corporation (CNN). Vertical line indicates when Donald Trump launched his presidential campaign. The horizontal scale, ranging from −4 to +4, indicates the number of weeks since Trump’s announcement. The bold red line labeled “crime + immig” indicates the number of immigration-related articles published in any of the seven newspapers that mentioned rape, drugs, or crime. Figure 2. View largeDownload slide Media coverage of immigration (May 19–July 13, 2015)Note: Lines show the total number of articles related to immigration published by each of the seven largest newspapers in the United States: the New York Times, Los Angeles Times, Washington Post, Wall Street Journal, Boston Globe, USA Today, and Orange County Register. It also includes news coverage data from all radio shows produced by National Public Radio (NPR) as well as transcripts from TV news and online articles produced by the Cable News Corporation (CNN). Vertical line indicates when Donald Trump launched his presidential campaign. The horizontal scale, ranging from −4 to +4, indicates the number of weeks since Trump’s announcement. The bold red line labeled “crime + immig” indicates the number of immigration-related articles published in any of the seven newspapers that mentioned rape, drugs, or crime. Further, I also find that the content of these news items changed by examining the articles published in these seven national newspapers. The red line indicates the number of all immigration-related articles in these newspapers that mentioned rape, drugs, or crime. It shows that a week before Trump’s speech, the number of newspaper articles linking immigrants to crime, which was the centerpiece of this politician’s announcement, doubled, going from 29 to 63. Such figure was again 63 one week later, and it increased to 143 and 202 articles two and three weeks after Trump’s announcement, respectively. This suggests that there could be two components to the treatment: the initial Trump speech as well as the subsequent media coverage and discussion triggered by the speech, which I will address in the second part of the article. The second condition I need to meet is that assignment to the treatment and control groups has to be random. Or, perhaps more realistically, there should be no major socio-demographic differences between both groups. If there were large compositional differences between these two groups, my design would be invalid, since these differences could be driving any perceived changes in attitudes. I use a relatively narrow window of time since there were plenty of respondents interviewed during those four days, but also because increasing the window period by adding more days would probably result in unobserved heterogeneity in the sample. In other words, survey takers interviewed on later dates may be increasingly different than the control group. However, in subsequent analyses I show the robustness of my results to the use of a wider time period. Gallup Survey Results Table 1 shows the descriptive statistics of the Gallup Minority Rights’ Survey for the analyzed period (June 15–18, 2015). This time period captures respondents interviewed two days after Trump’s speech and two days before. I use two-tailed two-sample t-tests assuming unequal variances to test whether there are any significant socio-demographic differences between respondents interviewed before Trump’s announcement and those interviewed after Trump’s announcement. I test a number of individual characteristics that could shape immigration attitudes, including age, gender, race, education, partisanship, community size, employment status, and geographic region. Table 1 shows that there are almost no statistically significant differences in the sample composition before and after Trump’s speech. Both samples are quite comparable in terms of their race, ethnicity, age, employment status, community type, and partisanship. The only exception is living in the Midwest or Western regions. The sample contains more Midwestern and fewer Western residents in the post-Trump period. I will return to this later on. In addition, a Gallup manager told me that the team in charge of this survey produced a sampling frame every day based on a national random sample of land- and cell-phone lines from across the country, which they call a “replicate.” This explains why there are no substantial socio-demographic differences in the sample composition by day, since every day Gallup composed a random national snapshot of US residents, weighted by key socio-demographic weights. This information corroborates that interviewing order was designed to be random. Table 1. Gallup Minority Rights’ Survey: Descriptive Statistics (n = 1,104) Mean/% Before Trump After Trump Race  White 64.5 64.8 64.3  African American 19.6 17.3 20.8  Other 15.7 17.8 14.8 Female 44.3 41.6 45.6 Hispanic 18.1 20.3 17.1 Age categories  18–29 23.1 22.8 23.4  30–49 40.3 40.1 40.4  50–64 56.6 56.9 56.6  65+ 74.1 74.1 74.1 Rural/small town 26.7 26.5 26.9 Unemployed 6.6 8.1 6.0 College 29.4 28.5 29.9 Geographic area  East 16.6 18.1 15.8  Midwest 16.5 11.0 19.2*  South 26.6 25.4 27.1  West 40.1 45.5 37.7* Partisanship  Democrat 37.1 36.0 37.7  Independent 39.4 43.2 37.7  Republican 20.8 18.4 21.9 Mean/% Before Trump After Trump Race  White 64.5 64.8 64.3  African American 19.6 17.3 20.8  Other 15.7 17.8 14.8 Female 44.3 41.6 45.6 Hispanic 18.1 20.3 17.1 Age categories  18–29 23.1 22.8 23.4  30–49 40.3 40.1 40.4  50–64 56.6 56.9 56.6  65+ 74.1 74.1 74.1 Rural/small town 26.7 26.5 26.9 Unemployed 6.6 8.1 6.0 College 29.4 28.5 29.9 Geographic area  East 16.6 18.1 15.8  Midwest 16.5 11.0 19.2*  South 26.6 25.4 27.1  West 40.1 45.5 37.7* Partisanship  Democrat 37.1 36.0 37.7  Independent 39.4 43.2 37.7  Republican 20.8 18.4 21.9 Source: Gallup Minority Rights’ Survey 2015. Descriptive statistics include respondents interviewed between June 15 and 18, 2015. “After Trump” indicates that the survey interview was conducted on either June 17 or 18 (after Donald Trump’s presidential campaign announcement). Two-tailed two-sample t-tests assuming unequal variances compare differences between respondents interviewed before and those reached after Trump’s announcement. *** p < 0.001 ** p < 0.01 * p < 0.05. Table 1. Gallup Minority Rights’ Survey: Descriptive Statistics (n = 1,104) Mean/% Before Trump After Trump Race  White 64.5 64.8 64.3  African American 19.6 17.3 20.8  Other 15.7 17.8 14.8 Female 44.3 41.6 45.6 Hispanic 18.1 20.3 17.1 Age categories  18–29 23.1 22.8 23.4  30–49 40.3 40.1 40.4  50–64 56.6 56.9 56.6  65+ 74.1 74.1 74.1 Rural/small town 26.7 26.5 26.9 Unemployed 6.6 8.1 6.0 College 29.4 28.5 29.9 Geographic area  East 16.6 18.1 15.8  Midwest 16.5 11.0 19.2*  South 26.6 25.4 27.1  West 40.1 45.5 37.7* Partisanship  Democrat 37.1 36.0 37.7  Independent 39.4 43.2 37.7  Republican 20.8 18.4 21.9 Mean/% Before Trump After Trump Race  White 64.5 64.8 64.3  African American 19.6 17.3 20.8  Other 15.7 17.8 14.8 Female 44.3 41.6 45.6 Hispanic 18.1 20.3 17.1 Age categories  18–29 23.1 22.8 23.4  30–49 40.3 40.1 40.4  50–64 56.6 56.9 56.6  65+ 74.1 74.1 74.1 Rural/small town 26.7 26.5 26.9 Unemployed 6.6 8.1 6.0 College 29.4 28.5 29.9 Geographic area  East 16.6 18.1 15.8  Midwest 16.5 11.0 19.2*  South 26.6 25.4 27.1  West 40.1 45.5 37.7* Partisanship  Democrat 37.1 36.0 37.7  Independent 39.4 43.2 37.7  Republican 20.8 18.4 21.9 Source: Gallup Minority Rights’ Survey 2015. Descriptive statistics include respondents interviewed between June 15 and 18, 2015. “After Trump” indicates that the survey interview was conducted on either June 17 or 18 (after Donald Trump’s presidential campaign announcement). Two-tailed two-sample t-tests assuming unequal variances compare differences between respondents interviewed before and those reached after Trump’s announcement. *** p < 0.001 ** p < 0.01 * p < 0.05. Having largely ruled out differences in sample composition, I now examine whether Trump’s speech affected public attitudes toward immigrants. As mentioned earlier, Gallup included the most commonly used question to measure immigration attitudes, which reads: “In your view, should immigration be kept at its present level, increased, or decreased?” In figure 3, I plot the responses to our main question by days since Trump’s speech. The vertical red line indicates day 0, which is when Trump gave his official speech. Day 1 represents the day after this speech, which is when most media outlets began reacting to his pronouncement. Figure 3. View largeDownload slide Individual preferences over immigration (N = 2,004) Source: Gallup’s Survey on Minority Rights and Relations 2015. Question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Vertical line indicates the day of Donald Trump’s presidential campaign announcement (June 16, 2015). Figure 3. View largeDownload slide Individual preferences over immigration (N = 2,004) Source: Gallup’s Survey on Minority Rights and Relations 2015. Question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Vertical line indicates the day of Donald Trump’s presidential campaign announcement (June 16, 2015). Figure 3 shows that public opinion on immigrants hardened after Trump’s speech. While 33 percent of respondents believed that immigration levels should be decreased a day after Trump’s speech, only 25 percent believed so a day before. Similarly, 23 percent of survey takers stated that immigration flows should be increased a day after the speech. Such number was 36 percent the day before. Figure 3 also shows a gradual hardening trend over this time period. For example, 10 days after Trump’s speech, 39 percent of respondents believed immigration levels should be decreased, which represents a 56 percent increase in this group relative to the day when Trump gave his speech. I now turn to regression analysis to examine these results in a more systematic manner. Since respondents could choose among three different options to express their opinion about current immigration flows (i.e., “decrease,” “increase,” and “keep the same”), I use multinomial regression models to predict their responses. The reference category for all models is “increase.” The key coefficient is AfterTrump, which indicates being interviewed after Trump’s speech. Therefore, this coefficient represents the effect of this speech on the individual’s immigration attitudes. Table 2 shows the results in relative risk ratios. All models include a dummy variable that indicates whether respondents were interviewed after Trump’s speech. In model 1, I compare respondents that were interviewed a day after Trump’s speech with those interviewed one day before. It shows that respondents interviewed after Trump’s speech had a higher relative risk to choose either “decrease” or “keep the same” relative to choosing “increase” than those interviewed before.6 These differences are statistically significant at the 99 percent level. In model 2, I test the robustness of this finding to expanding the time window to two days before and two days after the speech. While choosing “same” is no longer significantly different relative to choosing “increase,” respondents were still more likely to favor decreasing immigration flows even after expanding the time window. I consider this model to be the best specification since it maximizes the number of respondents in both control and treatment conditions but also keeps the time window relatively narrow, which minimizes concerns about selection (since respondent heterogeneity could increase as the time window expands). Table 2. Regression Models Predicting Individual Support for Immigration (Gallup Survey) Model 1: One day Model 2: Two days Model 3: Four days Model 4: Seven days Model 5: Is immigration bad? VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.802** 2.022** 1.279 1.407* 1.328+ 1.416* 1.244 1.425* 1.598* (0.396) (0.483) (0.200) (0.234) (0.202) (0.230) (0.177) (0.216) (0.365) Constant 1.052 0.701* 1.211 0.895 1.211 0.895 1.211 0.895 0.211*** (0.168) (0.125) (0.153) (0.122) (0.153) (0.122) (0.153) (0.122) (0.038) Observations 501 501 1,104 1,104 1,253 1,253 1,855 1,855 493 Model 1: One day Model 2: Two days Model 3: Four days Model 4: Seven days Model 5: Is immigration bad? VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.802** 2.022** 1.279 1.407* 1.328+ 1.416* 1.244 1.425* 1.598* (0.396) (0.483) (0.200) (0.234) (0.202) (0.230) (0.177) (0.216) (0.365) Constant 1.052 0.701* 1.211 0.895 1.211 0.895 1.211 0.895 0.211*** (0.168) (0.125) (0.153) (0.122) (0.153) (0.122) (0.153) (0.122) (0.038) Observations 501 501 1,104 1,104 1,253 1,253 1,855 1,855 493 Source: Gallup Minority Rights’ Survey 2015. Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 +p < 0.10. “After Trump” indicates that the survey interview was conducted after Donald Trump’s presidential campaign announcement. Analyzed question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Models 1–4 show results, in relative risk ratios, for multinomial logistic regression predicting responses to this question. Reference category for these multinomial regression models is “increased.” Model 1 includes all respondents interviewed the day of the speech and the day after. Model 2 includes all respondents interviewed two days before and two days after. Model 3 includes respondents interviewed two days before and four days after. Model 4 includes all respondents interviewed two days before and seven days later. The last column, model 5, analyzes the following question: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” This logistic regression predicts responding “bad thing” relative to responding “good thing” or “mixed.” Results for model 5 shown in odd ratios. Table 2. Regression Models Predicting Individual Support for Immigration (Gallup Survey) Model 1: One day Model 2: Two days Model 3: Four days Model 4: Seven days Model 5: Is immigration bad? VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.802** 2.022** 1.279 1.407* 1.328+ 1.416* 1.244 1.425* 1.598* (0.396) (0.483) (0.200) (0.234) (0.202) (0.230) (0.177) (0.216) (0.365) Constant 1.052 0.701* 1.211 0.895 1.211 0.895 1.211 0.895 0.211*** (0.168) (0.125) (0.153) (0.122) (0.153) (0.122) (0.153) (0.122) (0.038) Observations 501 501 1,104 1,104 1,253 1,253 1,855 1,855 493 Model 1: One day Model 2: Two days Model 3: Four days Model 4: Seven days Model 5: Is immigration bad? VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.802** 2.022** 1.279 1.407* 1.328+ 1.416* 1.244 1.425* 1.598* (0.396) (0.483) (0.200) (0.234) (0.202) (0.230) (0.177) (0.216) (0.365) Constant 1.052 0.701* 1.211 0.895 1.211 0.895 1.211 0.895 0.211*** (0.168) (0.125) (0.153) (0.122) (0.153) (0.122) (0.153) (0.122) (0.038) Observations 501 501 1,104 1,104 1,253 1,253 1,855 1,855 493 Source: Gallup Minority Rights’ Survey 2015. Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 +p < 0.10. “After Trump” indicates that the survey interview was conducted after Donald Trump’s presidential campaign announcement. Analyzed question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Models 1–4 show results, in relative risk ratios, for multinomial logistic regression predicting responses to this question. Reference category for these multinomial regression models is “increased.” Model 1 includes all respondents interviewed the day of the speech and the day after. Model 2 includes all respondents interviewed two days before and two days after. Model 3 includes respondents interviewed two days before and four days after. Model 4 includes all respondents interviewed two days before and seven days later. The last column, model 5, analyzes the following question: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” This logistic regression predicts responding “bad thing” relative to responding “good thing” or “mixed.” Results for model 5 shown in odd ratios. Model 3 is similar, but I expand the window of comparison to two days before and four days after the speech. I find that respondents interviewed after the speech were still more likely to choose “decrease” than those surveyed a day earlier. Similarly, model 4 shows that the result is robust to increasing the time window to respondents interviewed within a seven-day period after Trump’s speech. The fact that an effect can still be found even after increasing the time window may indicate that Trump’s speech may have resulted in a relatively durable hardening of public opinion toward immigrants. Alternatively, it could also mean that there was an ongoing hardening trend of public opinion during this time period, perhaps motivated by ongoing media coverage and discussion of Trump’s speech. I address this point using an original survey experiment in the second part of the article. If Trump’s speech hardened public opinion toward immigrants, we should expect other immigration-related questions in the same survey to be similarly affected. Fortunately, Gallup included another question on immigration in the same survey. This question read: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” Respondents could choose “bad thing,” “good thing,” or “mixed.” I use a logistic regression to analyze this question. This model predicts responding “bad thing” relative to responding “good thing” or “mixed.” I combined these two last categories because only nine respondents chose “mixed” during this time window. In model 5, shown in table 3, I compare respondents that were interviewed a day after Trump’s speech with those Gallup reached a day before. In accordance with our previous results, individuals interviewed right after Trump’s speech were more likely to view immigration as a “bad thing” than those interviewed earlier. Table 3. Multinomial Logistic Regressions Predicting Individual Support for Immigration (Gallup Survey) Model 1: Region Model 2: Demog Model 3: SES Model 4: Full VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.239 1.365+ 1.234 1.337+ 1.250 1.361+ 1.275 1.361+ (0.195) (0.230) (0.198) (0.235) (0.203) (0.245) (0.210) (0.251) Region (ref = East)  Midwest 1.369 1.515 1.227 1.306 1.219 1.162 1.290 1.257 (0.352) (0.431) (0.321) (0.379) (0.324) (0.349) (0.345) (0.390)  South 1.000 2.016** 0.989 2.051** 0.954 1.911* 0.982 1.893* (0.234) (0.502) (0.234) (0.522) (0.229) (0.497) (0.240) (0.511)  West 0.937 1.116 0.971 1.162 0.962 1.111 1.006 1.188 (0.194) (0.260) (0.207) (0.283) (0.209) (0.275) (0.226) (0.308)  Female 0.942 1.026 0.977 1.043 1.017 1.136 (0.145) (0.167) (0.153) (0.174) (0.162) (0.196)  Hispanic 0.670* 0.640* 0.643* 0.583* 0.645* 0.622* (0.133) (0.140) (0.132) (0.132) (0.137) (0.146) Race (ref = White)  Black 0.684+ 0.639* 0.680+ 0.603* 0.790 1.001 (0.134) (0.137) (0.137) (0.133) (0.171) (0.241)  Other 0.707 0.806 0.738 0.821 0.778 0.951 (0.152) (0.187) (0.163) (0.196) (0.176) (0.242) Age 1.007 1.021*** 1.006 1.022*** 1.008+ 1.024*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.005) Education (ref = less than HS)  HS 1.345 1.183 1.450 1.358 (0.569) (0.497) (0.620) (0.623)  Technical/trade 0.374 1.193 0.448 1.826 (0.271) (0.696) (0.334) (1.202)  Incomplete college 1.556 1.437 1.612 1.632 (0.650) (0.595) (0.683) (0.736)  Complete college 0.824 0.620 0.878 0.769 (0.347) (0.262) (0.376) (0.351)  Graduate school 0.891 0.478+ 0.921 0.583 (0.377) (0.206) (0.396) (0.270) Unemployed 0.501* 0.752 0.503* 0.781 (0.151) (0.236) (0.155) (0.254) Partisanship (ref = Republican)  Democrat 0.581* 0.226*** (0.140) (0.057)  Independent 1.049 0.694 (0.249) (0.168)  Other 0.993 0.996 (0.589) (0.599) Community type (ref = big city)  Small city 0.873 0.903 (0.206) (0.238)  Big city suburb 0.907 0.739 (0.201) (0.189)  Small city suburb 1.165 1.167 (0.427) (0.482)  Town 0.862 1.042 (0.243) (0.321)  Rural 0.844 1.044 (0.250) (0.317) Constant 1.208 0.673+ 1.125 0.282*** 1.078 0.329* 1.164 0.448 (0.239) (0.156) (0.358) (0.098) (0.553) (0.172) (0.664) (0.263) Observations 1,104 1,104 1,091 1,091 1,080 1,080 1,080 1,080 Pseudo R-squared 0.010 0.010 0.029 0.029 0.046 0.046 0.070 0.070 Model 1: Region Model 2: Demog Model 3: SES Model 4: Full VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.239 1.365+ 1.234 1.337+ 1.250 1.361+ 1.275 1.361+ (0.195) (0.230) (0.198) (0.235) (0.203) (0.245) (0.210) (0.251) Region (ref = East)  Midwest 1.369 1.515 1.227 1.306 1.219 1.162 1.290 1.257 (0.352) (0.431) (0.321) (0.379) (0.324) (0.349) (0.345) (0.390)  South 1.000 2.016** 0.989 2.051** 0.954 1.911* 0.982 1.893* (0.234) (0.502) (0.234) (0.522) (0.229) (0.497) (0.240) (0.511)  West 0.937 1.116 0.971 1.162 0.962 1.111 1.006 1.188 (0.194) (0.260) (0.207) (0.283) (0.209) (0.275) (0.226) (0.308)  Female 0.942 1.026 0.977 1.043 1.017 1.136 (0.145) (0.167) (0.153) (0.174) (0.162) (0.196)  Hispanic 0.670* 0.640* 0.643* 0.583* 0.645* 0.622* (0.133) (0.140) (0.132) (0.132) (0.137) (0.146) Race (ref = White)  Black 0.684+ 0.639* 0.680+ 0.603* 0.790 1.001 (0.134) (0.137) (0.137) (0.133) (0.171) (0.241)  Other 0.707 0.806 0.738 0.821 0.778 0.951 (0.152) (0.187) (0.163) (0.196) (0.176) (0.242) Age 1.007 1.021*** 1.006 1.022*** 1.008+ 1.024*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.005) Education (ref = less than HS)  HS 1.345 1.183 1.450 1.358 (0.569) (0.497) (0.620) (0.623)  Technical/trade 0.374 1.193 0.448 1.826 (0.271) (0.696) (0.334) (1.202)  Incomplete college 1.556 1.437 1.612 1.632 (0.650) (0.595) (0.683) (0.736)  Complete college 0.824 0.620 0.878 0.769 (0.347) (0.262) (0.376) (0.351)  Graduate school 0.891 0.478+ 0.921 0.583 (0.377) (0.206) (0.396) (0.270) Unemployed 0.501* 0.752 0.503* 0.781 (0.151) (0.236) (0.155) (0.254) Partisanship (ref = Republican)  Democrat 0.581* 0.226*** (0.140) (0.057)  Independent 1.049 0.694 (0.249) (0.168)  Other 0.993 0.996 (0.589) (0.599) Community type (ref = big city)  Small city 0.873 0.903 (0.206) (0.238)  Big city suburb 0.907 0.739 (0.201) (0.189)  Small city suburb 1.165 1.167 (0.427) (0.482)  Town 0.862 1.042 (0.243) (0.321)  Rural 0.844 1.044 (0.250) (0.317) Constant 1.208 0.673+ 1.125 0.282*** 1.078 0.329* 1.164 0.448 (0.239) (0.156) (0.358) (0.098) (0.553) (0.172) (0.664) (0.263) Observations 1,104 1,104 1,091 1,091 1,080 1,080 1,080 1,080 Pseudo R-squared 0.010 0.010 0.029 0.029 0.046 0.046 0.070 0.070 Source: Gallup Minority Rights’ Survey 2015. Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 +p < 0.10. “After Trump” indicates that the survey interview was conducted after Donald Trump’s presidential campaign announcement. Respondent sample for all models consists of individuals interviewed two days before Trump’s announcement and two days after. Analyzed question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Models show results for multinomial logistic regression predicting responses to this question. Reference category for all models is “increased.” Table 3. Multinomial Logistic Regressions Predicting Individual Support for Immigration (Gallup Survey) Model 1: Region Model 2: Demog Model 3: SES Model 4: Full VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.239 1.365+ 1.234 1.337+ 1.250 1.361+ 1.275 1.361+ (0.195) (0.230) (0.198) (0.235) (0.203) (0.245) (0.210) (0.251) Region (ref = East)  Midwest 1.369 1.515 1.227 1.306 1.219 1.162 1.290 1.257 (0.352) (0.431) (0.321) (0.379) (0.324) (0.349) (0.345) (0.390)  South 1.000 2.016** 0.989 2.051** 0.954 1.911* 0.982 1.893* (0.234) (0.502) (0.234) (0.522) (0.229) (0.497) (0.240) (0.511)  West 0.937 1.116 0.971 1.162 0.962 1.111 1.006 1.188 (0.194) (0.260) (0.207) (0.283) (0.209) (0.275) (0.226) (0.308)  Female 0.942 1.026 0.977 1.043 1.017 1.136 (0.145) (0.167) (0.153) (0.174) (0.162) (0.196)  Hispanic 0.670* 0.640* 0.643* 0.583* 0.645* 0.622* (0.133) (0.140) (0.132) (0.132) (0.137) (0.146) Race (ref = White)  Black 0.684+ 0.639* 0.680+ 0.603* 0.790 1.001 (0.134) (0.137) (0.137) (0.133) (0.171) (0.241)  Other 0.707 0.806 0.738 0.821 0.778 0.951 (0.152) (0.187) (0.163) (0.196) (0.176) (0.242) Age 1.007 1.021*** 1.006 1.022*** 1.008+ 1.024*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.005) Education (ref = less than HS)  HS 1.345 1.183 1.450 1.358 (0.569) (0.497) (0.620) (0.623)  Technical/trade 0.374 1.193 0.448 1.826 (0.271) (0.696) (0.334) (1.202)  Incomplete college 1.556 1.437 1.612 1.632 (0.650) (0.595) (0.683) (0.736)  Complete college 0.824 0.620 0.878 0.769 (0.347) (0.262) (0.376) (0.351)  Graduate school 0.891 0.478+ 0.921 0.583 (0.377) (0.206) (0.396) (0.270) Unemployed 0.501* 0.752 0.503* 0.781 (0.151) (0.236) (0.155) (0.254) Partisanship (ref = Republican)  Democrat 0.581* 0.226*** (0.140) (0.057)  Independent 1.049 0.694 (0.249) (0.168)  Other 0.993 0.996 (0.589) (0.599) Community type (ref = big city)  Small city 0.873 0.903 (0.206) (0.238)  Big city suburb 0.907 0.739 (0.201) (0.189)  Small city suburb 1.165 1.167 (0.427) (0.482)  Town 0.862 1.042 (0.243) (0.321)  Rural 0.844 1.044 (0.250) (0.317) Constant 1.208 0.673+ 1.125 0.282*** 1.078 0.329* 1.164 0.448 (0.239) (0.156) (0.358) (0.098) (0.553) (0.172) (0.664) (0.263) Observations 1,104 1,104 1,091 1,091 1,080 1,080 1,080 1,080 Pseudo R-squared 0.010 0.010 0.029 0.029 0.046 0.046 0.070 0.070 Model 1: Region Model 2: Demog Model 3: SES Model 4: Full VARIABLES Same Decrease Same Decrease Same Decrease Same Decrease After Trump 1.239 1.365+ 1.234 1.337+ 1.250 1.361+ 1.275 1.361+ (0.195) (0.230) (0.198) (0.235) (0.203) (0.245) (0.210) (0.251) Region (ref = East)  Midwest 1.369 1.515 1.227 1.306 1.219 1.162 1.290 1.257 (0.352) (0.431) (0.321) (0.379) (0.324) (0.349) (0.345) (0.390)  South 1.000 2.016** 0.989 2.051** 0.954 1.911* 0.982 1.893* (0.234) (0.502) (0.234) (0.522) (0.229) (0.497) (0.240) (0.511)  West 0.937 1.116 0.971 1.162 0.962 1.111 1.006 1.188 (0.194) (0.260) (0.207) (0.283) (0.209) (0.275) (0.226) (0.308)  Female 0.942 1.026 0.977 1.043 1.017 1.136 (0.145) (0.167) (0.153) (0.174) (0.162) (0.196)  Hispanic 0.670* 0.640* 0.643* 0.583* 0.645* 0.622* (0.133) (0.140) (0.132) (0.132) (0.137) (0.146) Race (ref = White)  Black 0.684+ 0.639* 0.680+ 0.603* 0.790 1.001 (0.134) (0.137) (0.137) (0.133) (0.171) (0.241)  Other 0.707 0.806 0.738 0.821 0.778 0.951 (0.152) (0.187) (0.163) (0.196) (0.176) (0.242) Age 1.007 1.021*** 1.006 1.022*** 1.008+ 1.024*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.005) Education (ref = less than HS)  HS 1.345 1.183 1.450 1.358 (0.569) (0.497) (0.620) (0.623)  Technical/trade 0.374 1.193 0.448 1.826 (0.271) (0.696) (0.334) (1.202)  Incomplete college 1.556 1.437 1.612 1.632 (0.650) (0.595) (0.683) (0.736)  Complete college 0.824 0.620 0.878 0.769 (0.347) (0.262) (0.376) (0.351)  Graduate school 0.891 0.478+ 0.921 0.583 (0.377) (0.206) (0.396) (0.270) Unemployed 0.501* 0.752 0.503* 0.781 (0.151) (0.236) (0.155) (0.254) Partisanship (ref = Republican)  Democrat 0.581* 0.226*** (0.140) (0.057)  Independent 1.049 0.694 (0.249) (0.168)  Other 0.993 0.996 (0.589) (0.599) Community type (ref = big city)  Small city 0.873 0.903 (0.206) (0.238)  Big city suburb 0.907 0.739 (0.201) (0.189)  Small city suburb 1.165 1.167 (0.427) (0.482)  Town 0.862 1.042 (0.243) (0.321)  Rural 0.844 1.044 (0.250) (0.317) Constant 1.208 0.673+ 1.125 0.282*** 1.078 0.329* 1.164 0.448 (0.239) (0.156) (0.358) (0.098) (0.553) (0.172) (0.664) (0.263) Observations 1,104 1,104 1,091 1,091 1,080 1,080 1,080 1,080 Pseudo R-squared 0.010 0.010 0.029 0.029 0.046 0.046 0.070 0.070 Source: Gallup Minority Rights’ Survey 2015. Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 +p < 0.10. “After Trump” indicates that the survey interview was conducted after Donald Trump’s presidential campaign announcement. Respondent sample for all models consists of individuals interviewed two days before Trump’s announcement and two days after. Analyzed question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Models show results for multinomial logistic regression predicting responses to this question. Reference category for all models is “increased.” Did the effect of Trump’s speech vary within certain subpopulations? It is plausible that this speech had different effects for several reasons. First, perhaps certain subgroups like Republicans or working-class respondents were more likely to watch the presidential announcement of a Republican candidate like Donald Trump. Second, as predicted by the framing literature (Druckman, Peterson, and Slothuus 2013), Trump’s message could have resonated more among partisan groups: individuals predisposed to holding restrictionist views on immigration. To test this, in models not shown, I explore subgroup differences in the effect of Donald Trump’s speech. Figure 4 shows the results of these models in graphic form. It shows the effect of Trump’s speech on the predicted probabilities of wanting to decrease immigration flows by respondents’ education, partisanship, and race. In accordance with prior framing studies, it shows that Republican respondents and those without a college education were more likely to choose “decrease” after Trump’s speech at statistically significant levels. In contrast, the responses of college-educated, Democrat, and independent voters did not change significantly. Non-Hispanic whites were also more likely to express a desire to reduce immigration after Trump’s announcement, though such difference was within the margin of error. This evidence indicates that the attitudinal effects of Trump’s speech were largely driven by attitudinal changes among groups that would seem especially receptive to his message including lower-educated and Republican respondents, who tend to have restrictionist views on immigration. Figure 4. View largeDownload slide Predicted probabilities of wanting to decrease immigration flows Source: Gallup’s Survey on Minority Rights and Relations 2015. Graph shows the predicted probabilities of wanting to “decrease” immigration flows based on multinomial logistic regression models predicting responses to this question: “In your view, should immigration be kept at its present level, increased, or decreased?” Reference category for these multinomial regression models is “kept same.” “After” indicates that the survey interview was conducted after Donald Trump’s presidential campaign announcement. Sample includes all respondents interviewed two days after the speech and two days before. Figure 4. View largeDownload slide Predicted probabilities of wanting to decrease immigration flows Source: Gallup’s Survey on Minority Rights and Relations 2015. Graph shows the predicted probabilities of wanting to “decrease” immigration flows based on multinomial logistic regression models predicting responses to this question: “In your view, should immigration be kept at its present level, increased, or decreased?” Reference category for these multinomial regression models is “kept same.” “After” indicates that the survey interview was conducted after Donald Trump’s presidential campaign announcement. Sample includes all respondents interviewed two days after the speech and two days before. Robustness checks As I showed earlier, individuals interviewed before Trump’s speech did not differ much from those interviewed after (see table 1). The one exception is that Midwest residents were more likely and Western residents less likely to be interviewed after Trump’s presidential announcement. Therefore, in this section I test whether results are robust to the inclusion of socio-demographic controls, including geographic residence. The regression form is identical to the first equation, except that I add a vector of individual characteristics including age, race, ethnicity, sex, region, partisanship, community type, education, and employment status. Table 3 reproduces the multinomial logistic regression model predicting attitudes toward immigration flows. The respondent sample for all models consists of individuals interviewed within two days before Trump’s announcement and two days after. Model 1 includes controls for geographic residence. It shows that even after including these controls, being interviewed after Trump’s speech is correlated with wanting to decrease immigration flows (though this is significant only at the 90 percent level). Model 2 tests the robustness of these results to adding gender, race, age, and ethnicity indicators. Despite these controls, results remain positive and statistically significant. In model 3, I include controls for educational attainment and employment. Even with these controls, the association between being interviewed after Trump’s speech and wanting to decrease immigration flows remains positive and statistically significant. Finally, in model 4, I include all controls from model 3 but also add controls for partisanship and community type. Results remain robust to this model specification as well. Effects on Other Policies To further test the robustness of these findings, I analyze responses to another set of policies: affirmative action policies targeting women and minorities. If responses to these policies also varied after Trump’s announcement, this could be evidence of selection bias in the sense that certain respondents, such as conservative individuals, could have been more likely to be interviewed after Trump’s speech. Respondents were asked if they favored or opposed these programs for women and for “racial minorities” separately. I use a logistic regression to predict opposition to these programs. I test the effect of Trump’s speech on these items using different time intervals: one day before and one day after the speech, two days before and two days after, and finally, two days before and four days after. Table 4 shows that respondents interviewed after Trump’s speech did not have more conservative views on these items than those interviewed before it. Therefore, these findings further assuage concerns about selection bias. Table 4. Logistic Regressions Predicting Opposition to Affirmative Action Policies M1: 1 day M2: 2 days M3: 4 days M4: 1 day M5: 2 days M6: 4 days VARIABLES Women Women Women Minorities Minorities Minorities After Trump 1.012 0.947 0.860 1.007 0.929 0.857 (0.204) (0.135) (0.120) (0.189) (0.126) (0.114) Constant 0.421*** 0.435*** 0.435*** 0.656** 0.593*** 0.593*** (0.064) (0.051) (0.051) (0.093) (0.066) (0.066) Observations 484 1,083 1,228 486 1,079 1,223 M1: 1 day M2: 2 days M3: 4 days M4: 1 day M5: 2 days M6: 4 days VARIABLES Women Women Women Minorities Minorities Minorities After Trump 1.012 0.947 0.860 1.007 0.929 0.857 (0.204) (0.135) (0.120) (0.189) (0.126) (0.114) Constant 0.421*** 0.435*** 0.435*** 0.656** 0.593*** 0.593*** (0.064) (0.051) (0.051) (0.093) (0.066) (0.066) Observations 484 1,083 1,228 486 1,079 1,223 Source: Gallup Minority Rights’ Survey 2015. Odd ratios reported. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 +p < 0.10. Question read: “Do you generally favor or oppose affirmative action programs for women?” Second question was identical except that it asked about “racial minorities.” Answers were recoded so that 1 indicated opposition and 0 support for these policies. Models 1 and 4 include all respondents interviewed the day of Trump’s speech and the day after. Models 2 and 5 include all respondents interviewed two days before and two days after. Models 3 and 6 include respondents interviewed two days before and four days after. Table 4. Logistic Regressions Predicting Opposition to Affirmative Action Policies M1: 1 day M2: 2 days M3: 4 days M4: 1 day M5: 2 days M6: 4 days VARIABLES Women Women Women Minorities Minorities Minorities After Trump 1.012 0.947 0.860 1.007 0.929 0.857 (0.204) (0.135) (0.120) (0.189) (0.126) (0.114) Constant 0.421*** 0.435*** 0.435*** 0.656** 0.593*** 0.593*** (0.064) (0.051) (0.051) (0.093) (0.066) (0.066) Observations 484 1,083 1,228 486 1,079 1,223 M1: 1 day M2: 2 days M3: 4 days M4: 1 day M5: 2 days M6: 4 days VARIABLES Women Women Women Minorities Minorities Minorities After Trump 1.012 0.947 0.860 1.007 0.929 0.857 (0.204) (0.135) (0.120) (0.189) (0.126) (0.114) Constant 0.421*** 0.435*** 0.435*** 0.656** 0.593*** 0.593*** (0.064) (0.051) (0.051) (0.093) (0.066) (0.066) Observations 484 1,083 1,228 486 1,079 1,223 Source: Gallup Minority Rights’ Survey 2015. Odd ratios reported. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 +p < 0.10. Question read: “Do you generally favor or oppose affirmative action programs for women?” Second question was identical except that it asked about “racial minorities.” Answers were recoded so that 1 indicated opposition and 0 support for these policies. Models 1 and 4 include all respondents interviewed the day of Trump’s speech and the day after. Models 2 and 5 include all respondents interviewed two days before and two days after. Models 3 and 6 include respondents interviewed two days before and four days after. More substantively, these findings also disprove the notion of spillover effects of negative messages toward immigrants. One hypothesis could be that being exposed to these negative messages may make respondents more punitive in general regardless of the policy under consideration. Nevertheless, this evidence suggests that immigration statements primarily affect attitudes toward immigrants. The lack of effect on affirmative action policies may stem from the fact that respondents may see these laws as primarily affecting African Americans. Taken together, these results provide suggestive evidence that Donald Trump’s public speech portraying immigrants as rapists and criminals shaped public attitudes toward immigrants. This research design has multiple strengths. First, these results are robust to multiple robustness checks. In addition, since the setup is based on a real-life political event, it is highly realistic. Finally, the data set used is nationally representative, which assuages concerns about external validity. Nevertheless, the research design also has some limitations. First, figure 3 shows that after Trump’s speech, there was a gradual hardening of public opinion toward immigrants during the month of July. How can we explain this? Did Trump’s speech result in a permanent shift of attitudes? An alternative explanation is that Trump’s announcement had only a short-term effect but its impact was extended and perhaps intensified by ongoing media coverage of his speech. Media coverage has been shown to shape public opinion (Terkildsen and Schnell 1997). After all, as I show in figure 2, media attention to immigration increased and also became more negative after Trump’s speech. Unfortunately, this research design cannot disentangle these two processes. In addition, this analysis is based on the plausible but untested assumption that individuals were exposed to Trump’s speech, but this may not necessarily be the case (Barabas and Jerit 2010). Further, even if Trump’s speech was consequential, we are uncertain whether it was impactful because a recognized political actor gave it as part of a presidential campaign or whether the message would have been equally impactful had it been uttered by anyone else given the same level of media attention. Finally, since Trump’s statement toward immigrants was negative, we are unable to examine the counterfactual: whether a positive message would have been equally consequential. Panel Survey Experiment To address these lingering questions, I conducted an original national panel survey experiment in which individuals were randomly exposed to different immigration-related messages uttered by both politicians and non-politicians. In this experiment, respondents were asked to read a brief newspaper article containing a statement regarding immigration, which was experimentally manipulated to vary in its overall sentiment and in its author. Further, to test whether messages’ effects are durable or transitory, respondents were re-interviewed a few weeks after initially completing the survey. This experiment addresses several limitations of the initial analysis based on Gallup survey data. First, by relying on random assignment, it provides unbiased causal estimates of the power of politicians’ messages to shape public attitudes toward immigrants. Second, by testing whether the author of the message (politician or non-politician) is consequential, it is able to directly test the main hypothesis and disentangle the effects of messages sent by politicians from other messages that individuals may be exposed to in their daily lives. Third, since the panel experiment re-interviews respondents a few weeks after the initial interview, it is able to test whether the effects of being exposed to immigration messages are durable or transitory, which is a common critique of standard one-shot experimental research designs. The survey experiment, which was approved by the Institutional Review Board of the University of Michigan (HUM00113743), was implemented in spring 2016. Before I began collecting data, I registered the research design and pre-analysis plan at Evidence in Governance and Politics (EGAP), an online repository of social science experiments and observational studies (ID 20160318AA). In all, 1,430 adult US residents were recruited via Amazon’s Mechanical Turk Internet service, which is a website where users perform tasks like responding to online surveys for small payments. Ninety-two percent of respondents, or 1,317 individuals, finished the survey. Average survey completion time was four minutes. Each participant was paid 32 cents. The sample is not strictly representative of the US population. However, table 5 shows that it provides significantly more variation in terms of race, ethnicity, gender, age, geography, and socio-economic background than traditional experimental studies conducted on college students. Further, recent evidence suggests that national samples of online respondents provide very similar results to samples that are explicitly designed to be representative (Weinberg, Freese, and McElhattan 2014). Table 5. Survey Experiment: Descriptive Statistics (n = 1,317) Mean/% (SD) Min Max Race/ethnicity  White 75.17  African American 6.45  Asian 7.29  Hispanic 7.90  Other 3.19 Age 34.37 (11.19) 18 75 Community size 2.18 (0.91) 1 4 Education 4.1 (1.33) 1 8 State  California 11.0  Florida 7.0  Texas 7.0  New York 6.0  Pennsylvania 6.0  Ohio 5.0  Michigan 4.0  Illinois 4.0  North Carolina 3.0  Georgia 3.0  New Jersey 3.0  Indiana 3.0  Virginia 2.0  Washington 3.0  Other States 34.0 Mean/% (SD) Min Max Race/ethnicity  White 75.17  African American 6.45  Asian 7.29  Hispanic 7.90  Other 3.19 Age 34.37 (11.19) 18 75 Community size 2.18 (0.91) 1 4 Education 4.1 (1.33) 1 8 State  California 11.0  Florida 7.0  Texas 7.0  New York 6.0  Pennsylvania 6.0  Ohio 5.0  Michigan 4.0  Illinois 4.0  North Carolina 3.0  Georgia 3.0  New Jersey 3.0  Indiana 3.0  Virginia 2.0  Washington 3.0  Other States 34.0 Note: Community size 1 = large, 2 = medium-size city, 3 = small town, and 4 = rural. The education categories are 1 = less than HS, 2 = HS, 3 = some college, 4 = 2-year college, 5 = 4-year college, 6 = master’s, 7 = professional degree (i.e., MBA, JD, etc.), and 8 = PhD. Table 5. Survey Experiment: Descriptive Statistics (n = 1,317) Mean/% (SD) Min Max Race/ethnicity  White 75.17  African American 6.45  Asian 7.29  Hispanic 7.90  Other 3.19 Age 34.37 (11.19) 18 75 Community size 2.18 (0.91) 1 4 Education 4.1 (1.33) 1 8 State  California 11.0  Florida 7.0  Texas 7.0  New York 6.0  Pennsylvania 6.0  Ohio 5.0  Michigan 4.0  Illinois 4.0  North Carolina 3.0  Georgia 3.0  New Jersey 3.0  Indiana 3.0  Virginia 2.0  Washington 3.0  Other States 34.0 Mean/% (SD) Min Max Race/ethnicity  White 75.17  African American 6.45  Asian 7.29  Hispanic 7.90  Other 3.19 Age 34.37 (11.19) 18 75 Community size 2.18 (0.91) 1 4 Education 4.1 (1.33) 1 8 State  California 11.0  Florida 7.0  Texas 7.0  New York 6.0  Pennsylvania 6.0  Ohio 5.0  Michigan 4.0  Illinois 4.0  North Carolina 3.0  Georgia 3.0  New Jersey 3.0  Indiana 3.0  Virginia 2.0  Washington 3.0  Other States 34.0 Note: Community size 1 = large, 2 = medium-size city, 3 = small town, and 4 = rural. The education categories are 1 = less than HS, 2 = HS, 3 = some college, 4 = 2-year college, 5 = 4-year college, 6 = master’s, 7 = professional degree (i.e., MBA, JD, etc.), and 8 = PhD. Respondents were asked to read a brief newspaper article containing a statement on immigration. The author (politician or non-politician) and the statements’ overall sentiment (neutral, pro-, or anti-immigrant) were randomly manipulated. The experiment had a 3 × 2 between-subjects design presented in table 6. There were four treatment conditions plus a control: (1) positive statement about immigrants by local resident; (2) positive statement about immigrants by politician; (3) negative statement about immigrants by local resident; (4) negative statement about immigrants by politician; and (5) neutral statement about immigrants (control condition: no speaker identified). Table 6. Experimental Design (n = 1,317) Immigration stance Pro-immigrant Anti-immigrant Politician Condition #1 (270) Condition #2 (256) Non-politician Condition #3 (276) Condition #4 (256) Immigration stance Pro-immigrant Anti-immigrant Politician Condition #1 (270) Condition #2 (256) Non-politician Condition #3 (276) Condition #4 (256) Note: Table shows the number of respondents on each experimental condition. A fifth condition, the control group, had 259 respondents. Table 6. Experimental Design (n = 1,317) Immigration stance Pro-immigrant Anti-immigrant Politician Condition #1 (270) Condition #2 (256) Non-politician Condition #3 (276) Condition #4 (256) Immigration stance Pro-immigrant Anti-immigrant Politician Condition #1 (270) Condition #2 (256) Non-politician Condition #3 (276) Condition #4 (256) Note: Table shows the number of respondents on each experimental condition. A fifth condition, the control group, had 259 respondents. To increase the realism of the treatment, the message was constructed based on real statements uttered by several high-profile restrictionist politicians, including Donald Trump, Pennsylvania congressman Lou Barletta, and former Arizona governor Jan Brewer. To preclude the identification of any of these figures based on this message, their statements were thoroughly blended and reworded to preserve their spirit but without making their origin too obvious. Pretesting was done via Amazon’s Mechanical Turk to verify that respondents could not identify the author of the statement. In the message, the speaker lists what he perceives are the negative consequences of immigration, including crime and increased health care and education costs. In the positive version, the same message is modified to change the sentiment of the message while keeping its structure and length intact. For example, while the negative version of the message ends with “we cannot stand idly by as these criminals compromise our quality of life,” the positive one does so with “we should support these immigrants since they improve our quality of life.” For experimental reasons, the messages were designed to be mirror images of each other. However, it is possible that this reduced the positivity of the positive message, since it included the same references as the negative message to immigration-related costs even if it questioned them. Finally, a control condition was added in which immigration was primed in a neutral way. This condition described how the US Census defines immigration. All treatment conditions had a similar word length—between 87 and 96 words. See Appendix A for a full description of all conditions. Past research shows that messengers’ traits may affect the attitudinal effects of elite cues (Kuklinski and Hurley 1994). Controversial messengers like Dr. Kevorkian or the KKK may weaken the impact of the message on opinion (Joslyn and Haider-Markel 2006; Nicholson 2011). In this case, the controversial status of Donald Trump could have depressed the effect of its immigration message among liberals but magnified it among conservatives. To test this, the identity of the speaker was randomly manipulated to include a non-controversial politician and, to determine whether the messenger’s status matters, a non-politician local resident. Some respondents were told that Wyoming governor Matt Mead had uttered the message, and others were told that a “local resident” named Matt Reed had made it. Governor Mead, a Republican, was chosen because Wyoming is the least populous state in the United States.7 As such, relatively few people would know Governor Mead’s actual position on immigration. This would make his utterance of either negative or positive statements about migration more credible than if we had used a more popular governor whose positions were better known. A randomization check is included in Appendix A. In addition, to test whether the effects of politicians’ statements are durable or ephemeral, respondents were randomly assigned to one of two experimental conditions: those contacted to take a follow-up survey after two weeks and those contacted after four weeks. Six hundred thirty-eight respondents were asked to take the survey after two weeks, and 454 of them (or 71 percent) completed the follow-up questionnaire. Six hundred thirty-one individuals were asked to be re-interviewed after four weeks, and 421 individuals (66.6 percent) agreed to do so. Though there could be concerns that these attrition rates could bias results, I find that there are no systematic socio-demographic differences between individuals who agreed to take the follow-up survey and those who did not (see table A2 in the Appendix). The only significant difference is that older respondents were more likely to take the follow-up survey. In models not shown, I include age controls but results do not change substantively. The follow-up surveys did not contain any experimental treatments. Instead, to test whether the statements respondents read during the baseline survey were still consequential, the follow-up surveys included the same questions about immigration. To verify the validity of the experimental treatments, I conducted a series of manipulation checks after the application of the experiment.8 I found that the different experimental conditions successfully conveyed the treatment cues I intended. Experimental Results In the survey experiment, I included the exact same attitudinal question asked in the Gallup survey, which read: “In your view, should immigration be kept at its present level, increased, or decreased?” Mirroring the prior analysis, I use a multinomial logistic regression to test whether the experimental treatments affected individuals’ expressed desire to decrease immigration flows. In these models, shown in table 7, I include controls for pre-treatment covariates like race/ethnicity, sex, age, and partisanship. This should tighten the estimates of treatment effects without increasing any bias because treatment assignment is unrelated to these traits9 (Gelman and Hill 2006). Table 7. Regression Models Predicting Immigration Attitudes (Survey Experiment) Change immigration flows? Is immigration good? VARIABLES Same Decrease Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.809 0.883 (0.307) (0.378) (0.579) (0.168)  Local negative 1.195 2.255** 2.016* 1.127 (0.295) (0.631) (0.664) (0.213)  Politician positive 1.347 1.451 1.307 0.867 (0.317) (0.409) (0.440) (0.163)  Politician negative 1.098 2.455** 2.676** 1.360 (0.271) (0.682) (0.848) (0.263) Race/ethnicity  Non-Hispanic Black 1.499 1.363 1.649 1.696* (0.516) (0.530) (0.630) (0.440)  Asian 0.984 0.590 0.547 0.822 (0.275) (0.214) (0.252) (0.191)  Hispanic 1.504 0.908 0.839 0.844 (0.459) (0.323) (0.298) (0.196)  Other 0.516 0.439+ 0.386 0.980 (0.194) (0.218) (0.295) (0.322) Conservative/moderate 1.736* 6.679*** 3.649*** 2.616*** (0.434) (1.688) (0.780) (0.393) College 0.980 0.573** 0.488*** 0.603*** (0.156) (0.103) (0.0931) (0.0739) Female 0.865 1.066 1.232 1.246 (0.141) (0.194) (0.238) (0.153) Age 1.020* 1.031*** 1.002 0.999 (0.008) (0.009) (0.008) (0.005) Constant 1.337 0.339** 0.132*** 0.891 (0.429) (0.124) (0.0538) (0.221) Observations 1,317 1,317 1,317 1,317 Pseudo R-squared 0.078 0.078 0.044 0.044 Change immigration flows? Is immigration good? VARIABLES Same Decrease Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.809 0.883 (0.307) (0.378) (0.579) (0.168)  Local negative 1.195 2.255** 2.016* 1.127 (0.295) (0.631) (0.664) (0.213)  Politician positive 1.347 1.451 1.307 0.867 (0.317) (0.409) (0.440) (0.163)  Politician negative 1.098 2.455** 2.676** 1.360 (0.271) (0.682) (0.848) (0.263) Race/ethnicity  Non-Hispanic Black 1.499 1.363 1.649 1.696* (0.516) (0.530) (0.630) (0.440)  Asian 0.984 0.590 0.547 0.822 (0.275) (0.214) (0.252) (0.191)  Hispanic 1.504 0.908 0.839 0.844 (0.459) (0.323) (0.298) (0.196)  Other 0.516 0.439+ 0.386 0.980 (0.194) (0.218) (0.295) (0.322) Conservative/moderate 1.736* 6.679*** 3.649*** 2.616*** (0.434) (1.688) (0.780) (0.393) College 0.980 0.573** 0.488*** 0.603*** (0.156) (0.103) (0.0931) (0.0739) Female 0.865 1.066 1.232 1.246 (0.141) (0.194) (0.238) (0.153) Age 1.020* 1.031*** 1.002 0.999 (0.008) (0.009) (0.008) (0.005) Constant 1.337 0.339** 0.132*** 0.891 (0.429) (0.124) (0.0538) (0.221) Observations 1,317 1,317 1,317 1,317 Pseudo R-squared 0.078 0.078 0.044 0.044 Note: Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. For the first model, respondents were asked: “In your view, should immigration be kept at its present level, increased, or decreased?” Reference category was “increased.” In the second question, respondents were asked: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” Response categories were: “good thing,” “bad thing,” and “mixed.” The reference category in the regression model was “good.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. Table 7. Regression Models Predicting Immigration Attitudes (Survey Experiment) Change immigration flows? Is immigration good? VARIABLES Same Decrease Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.809 0.883 (0.307) (0.378) (0.579) (0.168)  Local negative 1.195 2.255** 2.016* 1.127 (0.295) (0.631) (0.664) (0.213)  Politician positive 1.347 1.451 1.307 0.867 (0.317) (0.409) (0.440) (0.163)  Politician negative 1.098 2.455** 2.676** 1.360 (0.271) (0.682) (0.848) (0.263) Race/ethnicity  Non-Hispanic Black 1.499 1.363 1.649 1.696* (0.516) (0.530) (0.630) (0.440)  Asian 0.984 0.590 0.547 0.822 (0.275) (0.214) (0.252) (0.191)  Hispanic 1.504 0.908 0.839 0.844 (0.459) (0.323) (0.298) (0.196)  Other 0.516 0.439+ 0.386 0.980 (0.194) (0.218) (0.295) (0.322) Conservative/moderate 1.736* 6.679*** 3.649*** 2.616*** (0.434) (1.688) (0.780) (0.393) College 0.980 0.573** 0.488*** 0.603*** (0.156) (0.103) (0.0931) (0.0739) Female 0.865 1.066 1.232 1.246 (0.141) (0.194) (0.238) (0.153) Age 1.020* 1.031*** 1.002 0.999 (0.008) (0.009) (0.008) (0.005) Constant 1.337 0.339** 0.132*** 0.891 (0.429) (0.124) (0.0538) (0.221) Observations 1,317 1,317 1,317 1,317 Pseudo R-squared 0.078 0.078 0.044 0.044 Change immigration flows? Is immigration good? VARIABLES Same Decrease Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.809 0.883 (0.307) (0.378) (0.579) (0.168)  Local negative 1.195 2.255** 2.016* 1.127 (0.295) (0.631) (0.664) (0.213)  Politician positive 1.347 1.451 1.307 0.867 (0.317) (0.409) (0.440) (0.163)  Politician negative 1.098 2.455** 2.676** 1.360 (0.271) (0.682) (0.848) (0.263) Race/ethnicity  Non-Hispanic Black 1.499 1.363 1.649 1.696* (0.516) (0.530) (0.630) (0.440)  Asian 0.984 0.590 0.547 0.822 (0.275) (0.214) (0.252) (0.191)  Hispanic 1.504 0.908 0.839 0.844 (0.459) (0.323) (0.298) (0.196)  Other 0.516 0.439+ 0.386 0.980 (0.194) (0.218) (0.295) (0.322) Conservative/moderate 1.736* 6.679*** 3.649*** 2.616*** (0.434) (1.688) (0.780) (0.393) College 0.980 0.573** 0.488*** 0.603*** (0.156) (0.103) (0.0931) (0.0739) Female 0.865 1.066 1.232 1.246 (0.141) (0.194) (0.238) (0.153) Age 1.020* 1.031*** 1.002 0.999 (0.008) (0.009) (0.008) (0.005) Constant 1.337 0.339** 0.132*** 0.891 (0.429) (0.124) (0.0538) (0.221) Observations 1,317 1,317 1,317 1,317 Pseudo R-squared 0.078 0.078 0.044 0.044 Note: Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. For the first model, respondents were asked: “In your view, should immigration be kept at its present level, increased, or decreased?” Reference category was “increased.” In the second question, respondents were asked: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” Response categories were: “good thing,” “bad thing,” and “mixed.” The reference category in the regression model was “good.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. Table 7 shows that individuals who read the negative statement about immigrants were more likely to state that they would like to decrease the current immigration levels in the United States. While individuals who were assigned to the control condition had a 23 percent predicted probability to choose “decrease,” individuals who read the negative statement about immigrants uttered by a local resident or a politician had a 37 or 41 percent predicted probability of doing so, respectively. These findings go in the same direction as those found using the Gallup survey. Surprisingly, altering the author of the statement did not make much difference. What mattered the most was the polarity of the message, but even then only negative statements were consequential. Positive messages about immigrants had no effect on immigration attitudes regardless of whether they were uttered by a politician or a local resident. Mirroring the analysis in the previous section, I conduct interaction tests by education, partisanship, and race. Just like I found then, non-Hispanic whites, conservatives, and lower-educated respondents were more likely to express a desire to decrease immigration flows when primed with negative framings of immigration.10 For example, while 46 percent of non-Hispanic whites wanted to decrease immigration after reading the governor’s negative statement on immigrants, such figure is only 23 percent for non-whites in the same condition (and 24 percent for the control group). With regard to educational attainment, while 50 percent of non-college-educated respondents chose to decrease immigration when assigned to the negative/politician condition, only 33 percent of college-educated survey takers preferred to decrease immigration flows after reading the negative statement from the governor. Finally, conservative respondents also seemed more responsive to negative messages about immigration than liberal individuals.11 Seventy-four percent of conservative respondents wanted to decrease immigration flows when exposed to the negative/governor condition (relative to 46 percent when assigned to the control condition). In contrast, only 29 percent of their liberal counterparts also chose to reduce immigration when assigned to the same condition (and 18 percent in the control group). Were attitudinal effects of negative statements durable or ephemeral? I test this by re-interviewing some respondents after two weeks and others after four and asking them the same attitudinal questions as in the baseline survey. Figure 5 provides a summary of these findings. Full regression results can be found in the Appendix. As figure 5 shows, I find no significant effects after two weeks. Similarly, among those respondents re-interviewed after four weeks, there were no significant differences in the likelihood to choose “decrease” across the different experimental conditions (figure 5).12 Figure 5. View largeDownload slide Predicted probabilities of wanting to decrease current immigration levels Note: Predicted probabilities were estimated based on results from a multinomial logistic regression. The question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Response options were: “Present level,” “increased,” and “decreased.” The reference category in the regression model was “increased.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. “2 weeks” indicates respondents that took a follow-up survey two weeks after the baseline survey. “4 weeks” indicates respondents that took this follow-up survey four weeks after the original survey. Figure 5. View largeDownload slide Predicted probabilities of wanting to decrease current immigration levels Note: Predicted probabilities were estimated based on results from a multinomial logistic regression. The question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Response options were: “Present level,” “increased,” and “decreased.” The reference category in the regression model was “increased.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. “2 weeks” indicates respondents that took a follow-up survey two weeks after the baseline survey. “4 weeks” indicates respondents that took this follow-up survey four weeks after the original survey. To verify the robustness of these findings, I analyze the effect of immigration statements on a second attitudinal item also included in the Gallup survey. Respondents were asked: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” Response categories were: “good thing,” “bad thing,” and “mixed.” As table 8 shows, respondents exposed to negative messages about immigrants were more likely to claim that immigrants were a “bad thing” relative to those assigned to the neutral condition. The effect was slightly larger among those who read the negative statement by a politician (15 percent) than among those who were told the negative statement came from a local resident (12 percent). In contrast, only 7 percent of individuals in the baseline condition stated that immigrants were a “bad thing.” Just like I found with the first item, no statistically significant effect was detected among those exposed to positive statements. Table 8. Multinomial Logistic Regressions Predicting Immigration Attitudes (Survey Experiment) Baseline 2 Weeks 4 Weeks VARIABLES Bad Mixed Bad Mixed Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.809 0.883 1.301 1.321 1.170 1.059 (0.579) (0.168) (0.760) (0.447) (0.646) (0.351)  Local negative 2.016* 1.127 1.670 0.910 0.912 1.811 (0.664) (0.213) (0.906) (0.309) (0.604) (0.591) Politician positive 1.307 0.867 0.852 1.025 1.506 1.747 (0.440) (0.163) (0.537) (0.358) (0.823) (0.602) Politician negative 2.676** 1.360 1.393 1.290 0.866 1.701 (0.848) (0.263) (0.785) (0.443) (0.534) (0.584) Constant 0.132*** 0.891 0.302 1.384 0.179* 0.457+ (0.053) (0.221) (0.226) (0.654) (0.135) (0.208) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.044 0.044 0.078 0.078 0.089 0.089 Baseline 2 Weeks 4 Weeks VARIABLES Bad Mixed Bad Mixed Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.809 0.883 1.301 1.321 1.170 1.059 (0.579) (0.168) (0.760) (0.447) (0.646) (0.351)  Local negative 2.016* 1.127 1.670 0.910 0.912 1.811 (0.664) (0.213) (0.906) (0.309) (0.604) (0.591) Politician positive 1.307 0.867 0.852 1.025 1.506 1.747 (0.440) (0.163) (0.537) (0.358) (0.823) (0.602) Politician negative 2.676** 1.360 1.393 1.290 0.866 1.701 (0.848) (0.263) (0.785) (0.443) (0.534) (0.584) Constant 0.132*** 0.891 0.302 1.384 0.179* 0.457+ (0.053) (0.221) (0.226) (0.654) (0.135) (0.208) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.044 0.044 0.078 0.078 0.089 0.089 Note: Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. Respondents were asked: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” Response categories were: “good thing,” “bad thing,” and “mixed.” The reference category in the regression model was “good.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. “2 weeks” indicates respondents that took a follow-up survey two weeks after the baseline survey. “4 weeks” indicates respondents that took this follow-up survey four weeks after the original survey. Models also include controls for partisanship, race/ethnicity, education, gender, and age. Table 8. Multinomial Logistic Regressions Predicting Immigration Attitudes (Survey Experiment) Baseline 2 Weeks 4 Weeks VARIABLES Bad Mixed Bad Mixed Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.809 0.883 1.301 1.321 1.170 1.059 (0.579) (0.168) (0.760) (0.447) (0.646) (0.351)  Local negative 2.016* 1.127 1.670 0.910 0.912 1.811 (0.664) (0.213) (0.906) (0.309) (0.604) (0.591) Politician positive 1.307 0.867 0.852 1.025 1.506 1.747 (0.440) (0.163) (0.537) (0.358) (0.823) (0.602) Politician negative 2.676** 1.360 1.393 1.290 0.866 1.701 (0.848) (0.263) (0.785) (0.443) (0.534) (0.584) Constant 0.132*** 0.891 0.302 1.384 0.179* 0.457+ (0.053) (0.221) (0.226) (0.654) (0.135) (0.208) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.044 0.044 0.078 0.078 0.089 0.089 Baseline 2 Weeks 4 Weeks VARIABLES Bad Mixed Bad Mixed Bad Mixed Treatment conditions (ref = neutral)  Local positive 1.809 0.883 1.301 1.321 1.170 1.059 (0.579) (0.168) (0.760) (0.447) (0.646) (0.351)  Local negative 2.016* 1.127 1.670 0.910 0.912 1.811 (0.664) (0.213) (0.906) (0.309) (0.604) (0.591) Politician positive 1.307 0.867 0.852 1.025 1.506 1.747 (0.440) (0.163) (0.537) (0.358) (0.823) (0.602) Politician negative 2.676** 1.360 1.393 1.290 0.866 1.701 (0.848) (0.263) (0.785) (0.443) (0.534) (0.584) Constant 0.132*** 0.891 0.302 1.384 0.179* 0.457+ (0.053) (0.221) (0.226) (0.654) (0.135) (0.208) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.044 0.044 0.078 0.078 0.089 0.089 Note: Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. Respondents were asked: “On the whole, do you think immigration is a good thing or a bad thing for this country today?” Response categories were: “good thing,” “bad thing,” and “mixed.” The reference category in the regression model was “good.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. “2 weeks” indicates respondents that took a follow-up survey two weeks after the baseline survey. “4 weeks” indicates respondents that took this follow-up survey four weeks after the original survey. Models also include controls for partisanship, race/ethnicity, education, gender, and age. Table 8 also shows that the findings for this survey item mirror my previous findings in another key respect: attitudinal effects dissipated in the follow-up surveys. No significant differences were found across treatment conditions for respondents interviewed either two or four weeks after the baseline survey. In the experiment, I also include three more items that seek to capture individual attitudes on three different policy domains: military service, death penalty, and fines against people who pollute the environment. See Appendix A for the wording of these items. Nevertheless, just like I found when analyzing the Gallup survey in the first part of the article, attitudinal effects of immigration statements were confined to immigration items. No spillover effect was detected. See table A4 in the Appendix for full results of reactions to these policies. In summary, the panel survey experiment confirmed that exposure to immigration-related messages can shape attitudes toward immigrants, especially among non-Hispanic whites, conservatives, and lower-educated respondents. Conclusions and Discussion Prior scholars have documented the connection between politicians’ statements and public opinion formation on values, candidates, and public policies. Less is known about politicians’ capacity to affect public attitudes toward groups they demonize, like immigrants, welfare recipients, and criminal offenders. I find that elites’ immigration-related statements have the power to shape public attitudes toward immigrants and identified three structural traits of their effects. First, these effects do not seem to be author-dependent. Second, they are circumscribed to negative statements, as positive ones are not consequential. Third, they are ephemeral. I use two types of methodologies that have unique strengths and complement each other. While the survey analysis is based on a nationally representative sample of the US population and is based on a real-life event, the survey experiment confirms this causal relationship and tests its durability. Politicians’ Statements Are Consequential, but May Not Be Author-Dependent In a previous ethnographic study, I found that the proposal of a high-profile anti-immigrant law by the mayor of Hazleton, PA, appeared to stir anti-immigrant feelings among the native population (Flores 2014). After interviewing more than a hundred Hazleton residents, I theorized that part of this effect could stem from politicians’ public statements. When promoting these laws, restrictionist politicians like the Hazleton mayor or Donald Trump accuse immigrants of abusing social services and other crimes. In that article, I theorized that such statements could harden public opinion toward immigrants (Flores 2014). In this article, I am able to document the independent power of politicians’ negative statements net of any other processes related to the passing of these policies. In light of prior research, it is somewhat surprising that Trump’s explicitly racial statements on immigrants had an attitudinal impact. According to Mendelberg (2001, 191), “the perception exists that for a candidate there is no quicker route to political suicide than being identified as a racist.” So why then were explicitly racial anti-immigrant statements effective? One possibility is that it is more socially acceptable to target racial and ethnic minorities when they are labeled as “immigrants.” Some individuals may believe that the term “immigrants” does not necessarily refer to a specific national-origin group, but it is ethnically neutral (even if Trump specifically referred to “Mexican” immigrants during his campaign). Further, immigrants’ disputed legality may allow political entrepreneurs and their followers to more openly express their racial resentment toward “immigrant” minority groups than toward “native” minorities without being as afraid of violating anti-racist social norms. Yet another possibility is that Trump’s ascendancy to power, based on an explicit rejection of “political correctness,” may usher in a new era in which openly racial language is no longer penalized and instead becomes a legitimate political tool. While I also find that the author of the messages does not seem very consequential, this could have been driven by the study’s limitations. One possibility is that statements by better-known or higher-ranking politicians may be more impactful. In this study, I use a relatively obscure governor to minimize the risk of confounders (i.e., perhaps using a more famous politician would confound the effect of her formal position with that of her personal reputation). However, this is a strict test given that part of the attitudinal power of political elites may be grounded on their fame. In Trump’s case, his fame may counteract the negative effect of his controversial status and produce attitudinal effects. At the same time, even if people accorded no special consideration to politicians’ statements, politicians’ declarations could still be highly consequential since they tend to be broadcasted in the media, as previous research has shown (Page, Shapiro, and Dempsey 1987). Polarity Matters Though prior researchers have mostly theorized about the effect of negative statements about immigration, in this study I go beyond these theoretical expectations and test whether positive messages could also affect public attitudes toward immigrants. I find that positive messages do not influence public attitudes toward immigrants. Why is that the case? One possibility is that these results are driven by social desirability bias. Though, as I suggest above, targeting immigrants may be more permissible than targeting native US minorities, it still may not be fully socially desirable. Nevertheless, upon hearing a public figure like a politician making an anti-immigrant statement, some of these individuals might feel emboldened to express their true, negative views. This would imply that their actual opinion about immigrants does not change. Rather, they become more willing to express their real views as the politician seemingly legitimates them. This “emboldening” theoretical model resonates with my prior work on the effect of punitive immigration laws on immigration attitudes (Flores 2014, 2017). I have found that the passage of these policies increased the willingness of local residents to express anti-immigrant views and become politically active to oppose immigrants. This increased level of anti-immigrant activism created the perception that natives’ views toward immigrants had hardened as a result of the policy even if their underlying immigration attitudes had remained unchanged (Flores 2014, 2017). This would also explain why positive messages about immigration failed to produce attitudinal effects, since pro-immigrant respondents may be less concerned about social desirability and, thus, may be unaffected by politician statements since they do not feel the need to conceal their true feelings. Since US national rhetoric posits that “we are a nation of immigrants” (Alba and Nee 2009), publicly opposing immigrants may be less socially desirable than embracing them. Effects Are Ephemeral This study also illuminates the temporal dimension of this process, as I find that the statements’ effect on public attitudes toward immigrants is short-lived. Social desirability bias may also explain the ephemerality of these effects. If targeting minorities and immigrants is somewhat controversial for some, hearing a political figure openly chastising them may enable some individuals to express their true opinions. However, after a while, this “social permission” may wear off and there would be a return to the status quo, in which such views may still be somewhat controversial. This finding resonates with the literature on political advertisements, which finds that ads by politicians have small, short-term effects on political preferences (Gerber et al. 2011; Hill et al. 2013). This suggests that politicians’ rhetoric shares a fundamental characteristic with political ads. At their very core, both are political statements. The temporality of these effects also resonates with my previous ethnographic study of Hazleton, PA. The proposal of an anti-immigrant policy by the local mayor, Lou Barletta, energized anti-immigrant mobilization by natives (Flores 2014). But these effects declined over time, particularly after Barletta was elected to Congress and local media published fewer stories in which he denounced immigrants (Flores 2014; Longazel 2016). This suggests that to keep the population in a state of excitement, anti-immigrant politicians need to constantly repeat their restrictionist messages to their like-minded bases. This fact has been recognized by Trump himself, who revealed in an interview how he energizes the crowds: “You know, if it gets a little boring, if I see people starting to sort of, maybe thinking about leaving, I can sort of tell the audience, I just say, ‘We will build the wall!’ and they go nuts” (Penzenstadler 2016). Some caveats are in order. Though I show that politicians’ statements about immigration may have short-term attitudinal consequences, the signals I employ use multiple threat dimensions, including crime and tax costs. It is not entirely clear that each threat dimension is equally consequential to immigration attitudes, which future research could investigate. As mentioned earlier, the survey item used by Gallup mentions both “immigrants” and “immigration” in the same question. Nevertheless, these terms may evoke different connotations in people’s minds: immigrant people, immigration as a process, or even immigration policies. This could impact people’s expressed attitudes. For example, given US ideology as a “nation of immigrants” (Alba and Nee 2009), it is possible that individuals have more positive views toward immigrants as a social group than toward immigration policies, which both liberal and conservative politicians routinely criticize (Massey and Sánchez 2010). Future scholars could examine these different connotations and their consequences. Further, the framings used to portray positive messages on immigration may have limited their attitudinal effects. The positive message used, which mentioned criminal and education costs, was designed to closely mirror the negative frames employed by anti-immigrant politicians. Nevertheless, the sole mention of these costs, even if the positive messages questioned them, may have reduced the attitudinal effects of positive messages by reminding survey takers of the likely costs associated with immigration. A different framing could have been used to portray immigration in a positive light, such as human rights (Bloemraad, Silva, and Voss 2016). These caveats aside, however, this paper provides strong evidence that politicians’ immigration-related statements can shape expressed public attitudes toward immigrants, at least in the short term. Footnotes 1 For example, Jeb Bush, long considered the front-runner Republican candidate, did not mention immigration at all during his campaign announcement speech in June 2016 (Bush 2015). 2 A parallel literature in sociology studies collective action frames and framing processes in relation to social movements (see Benford and Snow 2000;,Johnston 1995). 3 My expectation is that most individuals were exposed to Trump’s speech via coverage by newspapers, radio, or TV the day after the event (or at nighttime, which is outside of Gallup’s allotted time to conduct interviews). However, it is possible that some survey respondents may have been exposed to it the same day it took place, either through live TV coverage or by browsing the Internet. I would anticipate that people without regular occupations outside their home were probably more likely to have seen coverage of it the same day it took place. In analyses not shown, I exclude all respondents that were more likely to be home the day of Trump’s announcement, including retired, disabled, and unemployed individuals. In other words, I restrict my sample to individuals who were probably not home the day of the speech and, hence, were more likely to find out about Trump’s speech during the night of June 16 or the morning after, including students, part-time and full-time workers, and unemployed people looking for work. I find that immigration attitudes significantly hardened after Trump’s speech even among these respondents. In addition, in a different set of models, I use an even more conservative approach by excluding respondents interviewed on June 16, the day of the speech, but I continue to find that Republican and non-Hispanic white respondents were significantly more likely to want to “decrease” immigration flows in the post-speech period. 4 News coverage data from other conservative sources like Fox News were not available. Nevertheless, we would expect those sources to pay even more attention to a Republican candidate like Donald Trump. 5 A long-standing sociological literature shows how political actors like groups or politicians use emotions as tools to mobilize individuals (see Bail 2015; Polletta and Jasper 2001). In this case, the angry language used by Trump may have contributed to the viral spread of his message. 6 In models not shown, I found that the relative risk of respondents choosing “same” relative to “decrease” did not significantly change in the post-Trump period. In other words, change took place between “increase” and other categories. 7 Indeed, only one survey respondent was from Wyoming. 8 In these manipulation tests, I found that 98 percent of respondents could correctly remember the identity of the speaker when he was a politician and 82–83 percent could recall it when he was a “local resident.” In addition, 98–99 percent of respondents could correctly recall that the text’s topic was immigration and 96–97 percent correctly recalled the text’s sentiment (whether it was positive or negative). Therefore, I conclude that the different experimental conditions successfully conveyed the treatment cues I intended. 9 Effects do not significantly vary when control variables are not included in the models. 10 Though the results for these interaction tests ran in the same direction as in the survey analysis, only the race interactions were statistically significant at conventional levels, perhaps due to small sample sizes. 11 These respondents include self-identified conservatives and moderates, which accounted for 24.15 percent of the sample. 12 Among respondents re-interviewed four weeks after the baseline survey, 26 percent of individuals assigned to the control condition wanted to “decrease” immigration levels in the United States. Such number was 24 percent and 27 percent for individuals exposed to positive and negative statements by a local resident, respectively. With regard to individuals who read a statement by a politician, 22 percent and 29 percent of those who read positive and negative statements, respectively, also wanted to decrease immigration levels. These differences are not statistically significant at conventional levels. Appendix A Experimental Prompts and Survey Items In this Appendix, we present the prompts for our experimental conditions and then list the survey questions each respondent should answer. Please read the following statement carefully. You will be asked some questions about it afterwards. Experimental conditions (only one was randomly displayed) 1. Positive statements about immigrants from nonpolitician (87 words) During a recent public gathering, Matt Reed, a local resident, said: “The U.S. has become a magnet for many talented people. We have over 10 million undocumented immigrants living in the U.S. but we have the resources to sustain it. It costs us little in terms of health care, education, or incarceration expenses.” Moreover, “because these immigrants are mostly peaceful, our communities are welcoming them. They are bringing new foods. They’re law-abiding. We should support these immigrants since they improve our quality of life,” the resident said. 2. Positive statements about immigrants from politician (88 words) During a recent public gathering, Matt Mead, the governor of Wyoming, said: “The U.S. has become a magnet for many talented people. We have over 10 million undocumented immigrants living in the U.S. but we have the resources to sustain it. It costs us little in terms of health care, education, or incarceration expenses.” Moreover, “because these immigrants are mostly peaceful, our communities are welcoming them. They are bringing new foods. They’re law-abiding. We should support these immigrants since they improve our quality of life,” the governor said. 3. Negative statements about immigrants from nonpolitician (95 words) During a recent public gathering, Matt Reed, a local resident, said: “The U.S. has become a dumping ground for everybody else’s problems. We have over 10 million illegal immigrants living in the U.S. and we do not have the resources to sustain it. It costs us a tremendous amount of money in health care, education, and incarceration.” Moreover, “because of violent acts committed by illegal aliens, our communities are gripped by fear. They are bringing drugs. They’re bringing crime. We cannot stand idly by as these criminals compromise our quality of life,” the resident said. 4. Negative statements about immigrants from politician (96 words) During a recent public gathering, Matt Mead, the governor of Wyoming, said: “The U.S. has become a dumping ground for everybody else’s problems. We have over 10 million illegal immigrants living in the U.S. and we do not have the resources to sustain it. It costs us a tremendous amount of money in health care, education, and incarceration.” Moreover, “because of violent acts committed by illegal aliens, our communities are gripped by fear. They are bringing drugs. They’re bringing crime. We cannot stand idly by as these criminals compromise our quality of life,” the governor said. 5. Control group (89 words) Migration is typically defined as a permanent geographic or spatial change in usual residence from one defined area to another. International migration consists of people changing permanent residence across national boundaries. In reference to the national estimates, net international migration to the United States is the net of 1) migration to the United States and 2) permanent departure from the United States. Temporary international movement of U.S. citizens, including the Armed Forces and their dependents, is not included in calculations of net international migration trends and is treated separately. Thinking now about immigrants—that is, people who come from other countries to live here in the United States, In your view, should immigration be kept at its present level, increased, or decreased? Present level Increased Decreased On the whole, do you think immigration is a good thing or a bad thing for this country today? Good thing Bad thing Mixed Do you favor or oppose the death penalty for persons convicted of murder? Favor Oppose Don’t know Even though they are no longer drafted for military service, young men are still required by law to register for the draft when they become 18 years old. If a young man refuses to register for the draft, do you think he should be punished in any way? Yes No Don’t know Do you favor or oppose to give stiff fines for individuals that pollute the environment? Favor Oppose Don’t know Table A1. Randomization Check (n = 1,317), percent Conditions White Black Hispanic College+ Age 40+ Conservative Neutral (control) 76.0 6.1 7.0 44.4 29.3 18.5 Positive/resident 76.1 5.4 6.5 51.8 22.8 25.7 Negative/resident 75.4 5.0 7.8 52.7 26.6 26.9 Positive/politician 70.7 8.9 10.0 49.6 25.2 22.9 Negative/politician 77.7 6.6 8.2 52.0 27.0 24.1 Conditions White Black Hispanic College+ Age 40+ Conservative Neutral (control) 76.0 6.1 7.0 44.4 29.3 18.5 Positive/resident 76.1 5.4 6.5 51.8 22.8 25.7 Negative/resident 75.4 5.0 7.8 52.7 26.6 26.9 Positive/politician 70.7 8.9 10.0 49.6 25.2 22.9 Negative/politician 77.7 6.6 8.2 52.0 27.0 24.1 Note: “Conservative” refers to self-identified conservatives. “College+” refers to respondents with at least a completed college degree. “White” and “Black” refer to non-Hispanic self-identified whites and blacks, respectively. “Hispanic” refers to self-identified Hispanics who can be of any race. Table A1. Randomization Check (n = 1,317), percent Conditions White Black Hispanic College+ Age 40+ Conservative Neutral (control) 76.0 6.1 7.0 44.4 29.3 18.5 Positive/resident 76.1 5.4 6.5 51.8 22.8 25.7 Negative/resident 75.4 5.0 7.8 52.7 26.6 26.9 Positive/politician 70.7 8.9 10.0 49.6 25.2 22.9 Negative/politician 77.7 6.6 8.2 52.0 27.0 24.1 Conditions White Black Hispanic College+ Age 40+ Conservative Neutral (control) 76.0 6.1 7.0 44.4 29.3 18.5 Positive/resident 76.1 5.4 6.5 51.8 22.8 25.7 Negative/resident 75.4 5.0 7.8 52.7 26.6 26.9 Positive/politician 70.7 8.9 10.0 49.6 25.2 22.9 Negative/politician 77.7 6.6 8.2 52.0 27.0 24.1 Note: “Conservative” refers to self-identified conservatives. “College+” refers to respondents with at least a completed college degree. “White” and “Black” refer to non-Hispanic self-identified whites and blacks, respectively. “Hispanic” refers to self-identified Hispanics who can be of any race. Table A2. Logistic Regression Predicting Responding to Follow-Up Surveys (1) (2) VARIABLES 2 Weeks 4 Weeks Treatment condition (ref = control)  Local positive 0.200 −0.0461 (0.292) (0.273)  Local negative 0.296 −0.048 (0.303) (0.282)  Politician positive −0.053 −0.225 (0.287) (0.271)  Politician negative 0.162 0.000 (0.297) (0.284) Age 0.047*** 0.048*** (0.010) (0.010) Race/ethnicity  Non-Hispanic white −0.508 0.152 (0.352) (0.367)  Non-Hispanic black 0.181 0.256 (0.340) (0.369)  Asian −0.332 0.523 (0.320) (0.328)  Hispanic 0.755 −0.235 (0.662) (0.427)  Other 0.294 0.191 (0.188) (0.176) Conservative −0.194 −0.222 (0.216) (0.207) Community size (ref = large city)  Medium city 0.077 0.044 (0.229) (0.224)  Small town −0.164 −0.064 (0.275) (0.253)  Rural 0.322 0.602 (0.365) (0.371) Constant −0.862+ −0.985* (0.452) (0.422) Observations 638 632 (1) (2) VARIABLES 2 Weeks 4 Weeks Treatment condition (ref = control)  Local positive 0.200 −0.0461 (0.292) (0.273)  Local negative 0.296 −0.048 (0.303) (0.282)  Politician positive −0.053 −0.225 (0.287) (0.271)  Politician negative 0.162 0.000 (0.297) (0.284) Age 0.047*** 0.048*** (0.010) (0.010) Race/ethnicity  Non-Hispanic white −0.508 0.152 (0.352) (0.367)  Non-Hispanic black 0.181 0.256 (0.340) (0.369)  Asian −0.332 0.523 (0.320) (0.328)  Hispanic 0.755 −0.235 (0.662) (0.427)  Other 0.294 0.191 (0.188) (0.176) Conservative −0.194 −0.222 (0.216) (0.207) Community size (ref = large city)  Medium city 0.077 0.044 (0.229) (0.224)  Small town −0.164 −0.064 (0.275) (0.253)  Rural 0.322 0.602 (0.365) (0.371) Constant −0.862+ −0.985* (0.452) (0.422) Observations 638 632 Note: Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 Table A2. Logistic Regression Predicting Responding to Follow-Up Surveys (1) (2) VARIABLES 2 Weeks 4 Weeks Treatment condition (ref = control)  Local positive 0.200 −0.0461 (0.292) (0.273)  Local negative 0.296 −0.048 (0.303) (0.282)  Politician positive −0.053 −0.225 (0.287) (0.271)  Politician negative 0.162 0.000 (0.297) (0.284) Age 0.047*** 0.048*** (0.010) (0.010) Race/ethnicity  Non-Hispanic white −0.508 0.152 (0.352) (0.367)  Non-Hispanic black 0.181 0.256 (0.340) (0.369)  Asian −0.332 0.523 (0.320) (0.328)  Hispanic 0.755 −0.235 (0.662) (0.427)  Other 0.294 0.191 (0.188) (0.176) Conservative −0.194 −0.222 (0.216) (0.207) Community size (ref = large city)  Medium city 0.077 0.044 (0.229) (0.224)  Small town −0.164 −0.064 (0.275) (0.253)  Rural 0.322 0.602 (0.365) (0.371) Constant −0.862+ −0.985* (0.452) (0.422) Observations 638 632 (1) (2) VARIABLES 2 Weeks 4 Weeks Treatment condition (ref = control)  Local positive 0.200 −0.0461 (0.292) (0.273)  Local negative 0.296 −0.048 (0.303) (0.282)  Politician positive −0.053 −0.225 (0.287) (0.271)  Politician negative 0.162 0.000 (0.297) (0.284) Age 0.047*** 0.048*** (0.010) (0.010) Race/ethnicity  Non-Hispanic white −0.508 0.152 (0.352) (0.367)  Non-Hispanic black 0.181 0.256 (0.340) (0.369)  Asian −0.332 0.523 (0.320) (0.328)  Hispanic 0.755 −0.235 (0.662) (0.427)  Other 0.294 0.191 (0.188) (0.176) Conservative −0.194 −0.222 (0.216) (0.207) Community size (ref = large city)  Medium city 0.077 0.044 (0.229) (0.224)  Small town −0.164 −0.064 (0.275) (0.253)  Rural 0.322 0.602 (0.365) (0.371) Constant −0.862+ −0.985* (0.452) (0.422) Observations 638 632 Note: Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05 Table A3. Multinomial Regression Models Predicting Immigration Attitudes (Survey Experiment) Baseline 2 weeks 4 weeks VARIABLES Same Decrease Same Decrease Same Decrease Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.229 1.104 1.753 1.254 (0.307) (0.378) (0.548) (0.539) (0.668) (0.570)  Local negative 1.195 2.255** 1.266 0.859 2.743* 1.925 (0.295) (0.631) (0.541) (0.404) (1.164) (0.961)  Politician positive 1.347 1.451 0.978 0.668 2.647* 1.450 (0.317) (0.409) (0.402) (0.317) (1.083) (0.670)  Politician negative 1.098 2.455** 1.265 0.901 3.988** 2.828 (0.271) (0.682) (0.540) (0.428) (1.919) (1.511) Race/ethnicity  Non-Hispanic Black 1.499 1.363 0.987 0.369 3.094 3.598 (0.516) (0.530) (0.525) (0.251) (2.578) (3.215)  Asian 0.984 0.590 2.065 0.978 0.669 0.657 (0.275) (0.214) (0.945) (0.624) (0.330) (0.410)  Hispanic 1.504 0.908 2.323 1.072 0.565 0.694 (0.459) (0.323) (1.373) (0.765) (0.243) (0.351)  Other 0.516 0.439 0.725 0.631 0.869 1.300 (0.194) (0.218) (0.521) (0.528) (0.623) (1.076) Conservative/mod 1.736* 6.679*** 1.584 7.347*** 2.007+ 5.449*** (0.434) (1.688) (0.647) (3.022) (0.827) (2.330) College 0.980 0.573** 0.920 0.621 0.800 0.505* (0.156) (0.103) (0.251) (0.191) (0.224) (0.162) Female 0.865 1.066 0.733 0.856 1.161 0.831 (0.141) (0.194) (0.192) (0.256) (0.316) (0.263) Age 1.020* 1.031*** 1.044** 1.040* 1.017 1.030 (0.008) (0.009) (0.015) (0.017) (0.014) (0.015) Constant 1.337 0.339** 0.539 0.373 0.787 0.345 (0.429) (0.124) (0.334) (0.264) (0.439) (0.220) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.0780 0.0780 0.0924 0.0924 0.0696 0.0696 Baseline 2 weeks 4 weeks VARIABLES Same Decrease Same Decrease Same Decrease Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.229 1.104 1.753 1.254 (0.307) (0.378) (0.548) (0.539) (0.668) (0.570)  Local negative 1.195 2.255** 1.266 0.859 2.743* 1.925 (0.295) (0.631) (0.541) (0.404) (1.164) (0.961)  Politician positive 1.347 1.451 0.978 0.668 2.647* 1.450 (0.317) (0.409) (0.402) (0.317) (1.083) (0.670)  Politician negative 1.098 2.455** 1.265 0.901 3.988** 2.828 (0.271) (0.682) (0.540) (0.428) (1.919) (1.511) Race/ethnicity  Non-Hispanic Black 1.499 1.363 0.987 0.369 3.094 3.598 (0.516) (0.530) (0.525) (0.251) (2.578) (3.215)  Asian 0.984 0.590 2.065 0.978 0.669 0.657 (0.275) (0.214) (0.945) (0.624) (0.330) (0.410)  Hispanic 1.504 0.908 2.323 1.072 0.565 0.694 (0.459) (0.323) (1.373) (0.765) (0.243) (0.351)  Other 0.516 0.439 0.725 0.631 0.869 1.300 (0.194) (0.218) (0.521) (0.528) (0.623) (1.076) Conservative/mod 1.736* 6.679*** 1.584 7.347*** 2.007+ 5.449*** (0.434) (1.688) (0.647) (3.022) (0.827) (2.330) College 0.980 0.573** 0.920 0.621 0.800 0.505* (0.156) (0.103) (0.251) (0.191) (0.224) (0.162) Female 0.865 1.066 0.733 0.856 1.161 0.831 (0.141) (0.194) (0.192) (0.256) (0.316) (0.263) Age 1.020* 1.031*** 1.044** 1.040* 1.017 1.030 (0.008) (0.009) (0.015) (0.017) (0.014) (0.015) Constant 1.337 0.339** 0.539 0.373 0.787 0.345 (0.429) (0.124) (0.334) (0.264) (0.439) (0.220) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.0780 0.0780 0.0924 0.0924 0.0696 0.0696 Note: Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. The question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Response options were: “Present level,” “increased,” and “decreased.” The reference category in the regression model was “increased.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. “2 weeks” indicates respondents that took a follow-up survey two weeks after the baseline survey. “4 weeks” indicates respondents that took this follow-up survey four weeks after the original survey. Table A3. Multinomial Regression Models Predicting Immigration Attitudes (Survey Experiment) Baseline 2 weeks 4 weeks VARIABLES Same Decrease Same Decrease Same Decrease Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.229 1.104 1.753 1.254 (0.307) (0.378) (0.548) (0.539) (0.668) (0.570)  Local negative 1.195 2.255** 1.266 0.859 2.743* 1.925 (0.295) (0.631) (0.541) (0.404) (1.164) (0.961)  Politician positive 1.347 1.451 0.978 0.668 2.647* 1.450 (0.317) (0.409) (0.402) (0.317) (1.083) (0.670)  Politician negative 1.098 2.455** 1.265 0.901 3.988** 2.828 (0.271) (0.682) (0.540) (0.428) (1.919) (1.511) Race/ethnicity  Non-Hispanic Black 1.499 1.363 0.987 0.369 3.094 3.598 (0.516) (0.530) (0.525) (0.251) (2.578) (3.215)  Asian 0.984 0.590 2.065 0.978 0.669 0.657 (0.275) (0.214) (0.945) (0.624) (0.330) (0.410)  Hispanic 1.504 0.908 2.323 1.072 0.565 0.694 (0.459) (0.323) (1.373) (0.765) (0.243) (0.351)  Other 0.516 0.439 0.725 0.631 0.869 1.300 (0.194) (0.218) (0.521) (0.528) (0.623) (1.076) Conservative/mod 1.736* 6.679*** 1.584 7.347*** 2.007+ 5.449*** (0.434) (1.688) (0.647) (3.022) (0.827) (2.330) College 0.980 0.573** 0.920 0.621 0.800 0.505* (0.156) (0.103) (0.251) (0.191) (0.224) (0.162) Female 0.865 1.066 0.733 0.856 1.161 0.831 (0.141) (0.194) (0.192) (0.256) (0.316) (0.263) Age 1.020* 1.031*** 1.044** 1.040* 1.017 1.030 (0.008) (0.009) (0.015) (0.017) (0.014) (0.015) Constant 1.337 0.339** 0.539 0.373 0.787 0.345 (0.429) (0.124) (0.334) (0.264) (0.439) (0.220) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.0780 0.0780 0.0924 0.0924 0.0696 0.0696 Baseline 2 weeks 4 weeks VARIABLES Same Decrease Same Decrease Same Decrease Treatment conditions (ref = neutral)  Local positive 1.312 1.330 1.229 1.104 1.753 1.254 (0.307) (0.378) (0.548) (0.539) (0.668) (0.570)  Local negative 1.195 2.255** 1.266 0.859 2.743* 1.925 (0.295) (0.631) (0.541) (0.404) (1.164) (0.961)  Politician positive 1.347 1.451 0.978 0.668 2.647* 1.450 (0.317) (0.409) (0.402) (0.317) (1.083) (0.670)  Politician negative 1.098 2.455** 1.265 0.901 3.988** 2.828 (0.271) (0.682) (0.540) (0.428) (1.919) (1.511) Race/ethnicity  Non-Hispanic Black 1.499 1.363 0.987 0.369 3.094 3.598 (0.516) (0.530) (0.525) (0.251) (2.578) (3.215)  Asian 0.984 0.590 2.065 0.978 0.669 0.657 (0.275) (0.214) (0.945) (0.624) (0.330) (0.410)  Hispanic 1.504 0.908 2.323 1.072 0.565 0.694 (0.459) (0.323) (1.373) (0.765) (0.243) (0.351)  Other 0.516 0.439 0.725 0.631 0.869 1.300 (0.194) (0.218) (0.521) (0.528) (0.623) (1.076) Conservative/mod 1.736* 6.679*** 1.584 7.347*** 2.007+ 5.449*** (0.434) (1.688) (0.647) (3.022) (0.827) (2.330) College 0.980 0.573** 0.920 0.621 0.800 0.505* (0.156) (0.103) (0.251) (0.191) (0.224) (0.162) Female 0.865 1.066 0.733 0.856 1.161 0.831 (0.141) (0.194) (0.192) (0.256) (0.316) (0.263) Age 1.020* 1.031*** 1.044** 1.040* 1.017 1.030 (0.008) (0.009) (0.015) (0.017) (0.014) (0.015) Constant 1.337 0.339** 0.539 0.373 0.787 0.345 (0.429) (0.124) (0.334) (0.264) (0.439) (0.220) Observations 1,317 1,317 445 445 416 416 Pseudo R-squared 0.0780 0.0780 0.0924 0.0924 0.0696 0.0696 Note: Relative risk ratios shown. Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. The question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Response options were: “Present level,” “increased,” and “decreased.” The reference category in the regression model was “increased.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. “2 weeks” indicates respondents that took a follow-up survey two weeks after the baseline survey. “4 weeks” indicates respondents that took this follow-up survey four weeks after the original survey. Table A4. Logistic Regressions Predicting Responses to Unrelated Policies (Survey Experiment) VARIABLES Death Penalty Pollution Fines Military Service Treatment conditions (ref = neutral)  Local positive 0.309 −0.475 −0.002 (0.174) (0.244) (0.189)  Local negative 0.162 0.138 −0.096 (0.177) (0.275) (0.194)  Politician positive 0.193 −0.180 −0.170 (0.174) (0.256) (0.193)  Politician negative 0.162 0.022 −0.020 (0.177) (0.269) (0.193) Constant −0.178 1.959*** −0.842*** (0.125) (0.189) (0.135) Observations 1,317 1,317 1,317 VARIABLES Death Penalty Pollution Fines Military Service Treatment conditions (ref = neutral)  Local positive 0.309 −0.475 −0.002 (0.174) (0.244) (0.189)  Local negative 0.162 0.138 −0.096 (0.177) (0.275) (0.194)  Politician positive 0.193 −0.180 −0.170 (0.174) (0.256) (0.193)  Politician negative 0.162 0.022 −0.020 (0.177) (0.269) (0.193) Constant −0.178 1.959*** −0.842*** (0.125) (0.189) (0.135) Observations 1,317 1,317 1,317 Note: Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. The question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Response options were: “Present level,” “increased,” and “decreased.” The reference category in the regression model was “increased.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. Table A4. Logistic Regressions Predicting Responses to Unrelated Policies (Survey Experiment) VARIABLES Death Penalty Pollution Fines Military Service Treatment conditions (ref = neutral)  Local positive 0.309 −0.475 −0.002 (0.174) (0.244) (0.189)  Local negative 0.162 0.138 −0.096 (0.177) (0.275) (0.194)  Politician positive 0.193 −0.180 −0.170 (0.174) (0.256) (0.193)  Politician negative 0.162 0.022 −0.020 (0.177) (0.269) (0.193) Constant −0.178 1.959*** −0.842*** (0.125) (0.189) (0.135) Observations 1,317 1,317 1,317 VARIABLES Death Penalty Pollution Fines Military Service Treatment conditions (ref = neutral)  Local positive 0.309 −0.475 −0.002 (0.174) (0.244) (0.189)  Local negative 0.162 0.138 −0.096 (0.177) (0.275) (0.194)  Politician positive 0.193 −0.180 −0.170 (0.174) (0.256) (0.193)  Politician negative 0.162 0.022 −0.020 (0.177) (0.269) (0.193) Constant −0.178 1.959*** −0.842*** (0.125) (0.189) (0.135) Observations 1,317 1,317 1,317 Note: Robust standard errors in parentheses. *** p < 0.001 ** p < 0.01 * p < 0.05. The question read: “In your view, should immigration be kept at its present level, increased, or decreased?” Response options were: “Present level,” “increased,” and “decreased.” The reference category in the regression model was “increased.” In the “Control” condition, respondents read a neutral immigration message. In “local positive,” respondents read a pro-immigrant message by a local resident. In “local negative,” the local resident uttered a negative message. Respondents assigned to the “politician positive” condition read a pro-immigrant statement by a politician. Finally, those assigned to the “politician negative” message read an anti-immigrant statement by a politician. About the Author René D. Flores is a Donald D. Harrington Faculty Fellow at the University of Texas and the Neubauer Family Assistant Professor of Sociology at the University of Chicago. His primary research interests are in the fields of international migration, race and ethnicity, and social stratification. His work has appeared in the American Journal of Sociology, American Sociological Review, Social Forces, and Social Problems. References Alba , Richard , and Victor Nee . 2009 . Remaking the American Mainstream: Assimilation and Contemporary Immigration . Cambridge, MA : Harvard University Press . Arceneaux , Kevin . 2008 . “ Can Partisan Cues Diminish Democratic Accountability? ” Political Behavior 30 ( 2 ): 139 – 60 . Google Scholar CrossRef Search ADS Bandler , Aaron . 2016 . “Here’s a Full Timeline of Donald Trump’s Immigration Positions.” http://www.dailywire.com/news/8644/heres-full-timeline-donald-trumps-immigration-aaron-bandler, accessed June 6, 2017. Bail , Christopher A. 2015 . Terrified: How Anti-Muslim Fringe Organizations Became Mainstream . Princeton, NJ : Princeton University Press . Google Scholar CrossRef Search ADS Barabas , Jason , and Jennifer Jerit . 2010 . “ Are Survey Experiments Externally Valid? ” American Political Science Review 104 : 226 – 42 . Google Scholar CrossRef Search ADS Beckett , Katherine . 1997 . Making Crime Pay: Law and Order in Contemporary American Politics . New York : Oxford University Press . Benford , Robert D. , and David A. Snow . 2000 . “ Framing Processes and Social Movements: An Overview and Assessment .” Annual Review of Sociology 26 ( 1 ): 611 – 39 . Google Scholar CrossRef Search ADS Blinder , S. , R. Ford , and E. Ivarsflaten . 2013 . “ The Better Angels of Our Nature: How the Antiprejudice Norm Affects Policy and Party Preferences in Great Britain and Germany .” American Journal of Political Science 57 ( 4 ): 841 – 57 . Bloemraad , Irene , Fabiana Silva , and Kim Voss . 2016 . “ Rights, Economics, or Family? Frame Resonance, Political Ideology, and the Immigrant Rights Movement .” Social Forces 94 ( 4 ): 1647 – 74 . Google Scholar CrossRef Search ADS Bonilla-Silva , Eduardo . 2003 . “ ‘New Racism,’ Color-Blind Racism, and the Future of Whiteness in America .” White Out: The Continuing Significance of Racism , 271 – 84 . Brown , Hana E. 2013 . “ Race, Legality, and the Social Policy Consequences of Anti-Immigration Mobilization .” American Sociological Review 78 ( 2 ): 290 – 314 . Google Scholar CrossRef Search ADS Burke , Michael . 2016 . “Activity among White Supremacists Continues to Surge.” USA Today, June 16. Bush , Jeb . 2015 . Full text of Jeb Bush’s presidential announcement. http://www.politico.com/story/2015/06/jeb-bush-2016-announcement-full-text-119023, accessed December 17, 2016. Calavita , Kitty . 1996 . “ The New Politics of Immigration: Balanced-Budget Conservatism and the Symbolism of Proposition 187 .” Social Problems 43 : 284 – 305 . Google Scholar CrossRef Search ADS Carroll , Rory . 2016 . “‘You Were Born in a Taco Bell’: Trump’s Rhetoric Fuels School Bullies Across US.” The Guardian, June 9. Chavez , Leo R. 2001 . Covering Immigration: Popular Images and the Politics of the Nation . Berkeley : University of California Press . ——— . 2008 . The Latino Threat: Constructing Immigrants, Citizens, and the Nation . Redwood City, CA : Stanford University Press . Citrin , J. , B. Reingold , and D. P. Green. 1990 . “ American Identity and the Politics of Ethnic Change .” Journal of Politics 57 ( 4 ): 1124 – 54 . Google Scholar CrossRef Search ADS Confessore , Nicholas , and Karen Yourish . 2016 . “$2 Billion Worth of Free Media for Donald Trump.” New York Times, March 15. Druckman , James N. 2001 . “ On the Limits of Framing Effects: Who Can Frame? ” Journal of Politics 63 ( 4 ): 1041 – 66 . Google Scholar CrossRef Search ADS Druckman , James N. , and Kjersten R. Nelson . 2003 . “ Framing and Deliberation: How Citizens’ Conversations Limit Elite Influence .” American Journal of Political Science 47 ( 4 ): 729 – 45 . Google Scholar CrossRef Search ADS Druckman , J. N. , E. Peterson , and R. Slothuus . 2013 . “ How Elite Partisan Polarization Affects Public Opinion Formation .” American Political Science Review 107 ( 1 ): 57 – 79 . Google Scholar CrossRef Search ADS Easton , D. , and J. Dennis . 1969 . Children in the Political System . New York : McGraw-Hill . Edelman , Murray . 1977 . Political Language: Words That Succeed and Policies That Fail . New York : Academic Press . Ehrenfreund , Max . 2016 . “What’s Incredible About Republicans’ Views on Immigration Is How Much They’ve Changed.” Washington Post, February 3. Flores , René D. 2014 . “ In the Eye of the Storm: How Did Hazleton’s Restrictive Immigration Ordinance Affect Local Interethnic Relations? ” American Behavioral Scientist 58 ( 13 ): 1743 – 63 . Google Scholar CrossRef Search ADS ——— . 2015 . “ Taking the Law into Their Own Hands: Do Local Anti-Immigrant Ordinances Increase Gun Sales? ” Social Problems 62 ( 3 ): 363 – 90 . Google Scholar CrossRef Search ADS ——— . 2017 . “ Do Anti-Immigrant Laws Shape Public Sentiment?: A Study of Arizona’s SB 1070 Using Twitter Data .” American Journal of Sociology 123 ( 2 ): 333 – 84 . Google Scholar CrossRef Search ADS Fox , Lauren . 2015 . “Boston Brothers Held on High Bail in Alleged Hate Crime.” Boston Globe, November 26. Franke-Ruta , Garance . 2013 . “What You Need to Read in the RNC Election-Autopsy Report.” The Atlantic, March 18. Gelman , Andrew , and Jennifer Hill . 2006 . Data Analysis Using Regression and Multilevel/Hierarchical Models . Cambridge : Cambridge University Press . Google Scholar CrossRef Search ADS Gerber , Alan S. , James G. Gimpel , Donald P. Green , and Daron R. Shaw . 2011 . “ How Large and Long-Lasting Are the Persuasive Effects of Televised Campaign Ads? Results from a Randomized Field Experiment .” American Political Science Review 105 ( 1 ): 135 – 50 . Google Scholar CrossRef Search ADS Goren , Paul , Christopher M. Federico , and Miki Caul Kittilson . 2009 . “ Source Cues, Partisan Identities, and Political Value Expression .” American Journal of Political Science 53 ( 4 ): 805 – 20 . Google Scholar CrossRef Search ADS Haberman , Maggie . 2016 . “Donald Trump’s Immigration Message May Resound in New Hampshire.” New York Times, February 5. Haberman , Maggie , and Alexander Burns . 2016 . “Donald Trump’s Presidential Run Began in an Effort to Gain Stature.” New York Times, March 12. Hainmueller , Jens , and Daniel J. Hopkins . 2014 . “ Public Attitudes toward Immigration .” Annual Review of Political Science 17 : 225 – 49 . Google Scholar CrossRef Search ADS Hill , Seth J. , James Lo , Lynn Vavreck , and John Zaller . 2013 . “ How Quickly We Forget: The Duration of Persuasion Effects from Mass Communication .” Political Communication 30 ( 4 ): 521 – 47 . Google Scholar CrossRef Search ADS Hopkins , D. J. 2010 . “ Politicized Places: Explaining Where and When Immigrants Provoke Local Opposition .” American Political Science Review 104 ( 1 ): 40 – 60 . Google Scholar CrossRef Search ADS Hopkins , Daniel J. , Van C. Tran , and Abigail F. Williamson . 2014 . “ See No Spanish: Language, Local Context, and Attitudes toward Immigration .” Politics, Groups, and Identities 2 ( 1 ): 35 – 51 . Google Scholar CrossRef Search ADS Hyman , H. H. 1959 . Political Socialization . New York : Free Press . Johnston , Hank . 1995 . “ A Methodology for Frame Analysis: From Discourse to Cognitive Schemata .” Social Movements and Culture 4 : 2l7 – 46 . Joslyn , Mark R. , and Donald P. Haider-Markel . 2006 . “ Should We Really ‘Kill’ the Messenger? Framing Physician-Assisted Suicide and the Role of Messengers .” Political Communication 23 ( 1 ): 85 – 103 . Google Scholar CrossRef Search ADS Kam , Cindy D. 2005 . “ Who Toes the Party Line? Cues, Values, and Individual Differences .” Political Behavior 27 ( 2 ): 163 – 82 . Google Scholar CrossRef Search ADS Krieg , Gregory . 2016 . “14 of Trump’s Most Outrageous ‘Birther’ Claims—Half from after 2011.” http://www.cnn.com/2016/09/09/politics/donald-trump-birther/index.html, accessed June 6, 2017. Kuklinski , James H. , and Norman L. Hurley . 1994 . “ On Hearing and Interpreting Political Messages: A Cautionary Tale of Citizen Cue-Taking .” Journal of Politics 56 ( 3 ): 729 – 51 . Google Scholar CrossRef Search ADS Legewie , J. 2013 . “ Terrorist Events and Attitudes toward Immigrants: A Natural Experiment .” American Journal of Sociology 118 ( 5 ): 1199 – 1245 . Google Scholar CrossRef Search ADS Longazel , Jamie . 2016 . Undocumented Fears: Immigration and the Politics of Divide and Conquer in Hazleton, Pennsylvania . Philadelphia : Temple University Press . Marrow , Helen B. 2011 . New Destination Dreaming: Immigration, Race, and Legal Status in the Rural American South . Redwood City, CA : Stanford University Press . Massey , Douglas S. , and Karen A. Pren . 2012 . “ Origins of the New Latino Underclass .” Race and Social Problems 4 ( 1 ): 5 – 17 . Google Scholar CrossRef Search ADS PubMed Massey , Douglas , and Magaly Sánchez . 2010 . Brokered Boundaries: Creating Immigrant Identity in Anti-Immigrant Times . New York : Russell Sage Foundation . Mendelberg , Tali . 2001 . The Race Card: Campaign Strategy, Implicit Messages, and the Norm of Equality . Princeton, NJ : Princeton University Press . Google Scholar CrossRef Search ADS Nelson , Thomas E. , and Donald R. Kinder . 1996 . “ Issue Frames and Group-Centrism in American Public Opinion .” Journal of Politics 58( 4 ): 1055 – 78 . Google Scholar CrossRef Search ADS Nicholson , Stephen P. 2011 . “ Dominating Cues and the Limits of Elite Influence .” Journal of Politics 73 ( 4 ): 1165 – 77 . Google Scholar CrossRef Search ADS ——— . 2012 . “ Polarizing Cues .” American Journal of Political Science 56 ( 1 ): 52 – 66 . Google Scholar CrossRef Search ADS PubMed Page , Benjamin I. , Robert Y. Shapiro , and Glenn R. Dempsey . 1987 . “ What Moves Public Opinion? ” American Political Science Review 81 ( 1 ): 23 – 43 . Google Scholar CrossRef Search ADS Penzenstadler , Nick . 2016 . “Trump: When Audiences Get Bored I Use ‘The Wall’.” USA TODAY. https://www.usatoday.com/story/news/politics/onpolitics/2016/01/30/trump-when-audiences-get-bored-use-wall/79573388/. Polletta , Francesca , and James M. Jasper . 2001 . “ Collective Identity and Social Movements .” Annual Review of Sociology 27 ( 1 ): 283 – 305 . Google Scholar CrossRef Search ADS Republican National Committee . 2013 . Growth and Opportunity Project. July 16. Rove , Karl . 2013 . “More White Votes Alone Won’t Save the GOP: To Win the Presidency in 2016, the Party Needs to Do Better with Hispanics.” Wall Street Journal, June 26. Rubin , Jennier . 2013 . “GOP Autopsy Report Goes Bold.” Washington Post, March 18. Sanneh , Kelefa . 2015 . “A Serious Immigration Debate, Thanks to Donald Trump.” New Yorker, August 19. Santa Ana , Otto . 2002 . Brown Tide Rising: Metaphoric Representations of Latinos in Contemporary American Public Discourse . Austin : University of Texas Press . Schachter , Ariela . 2015 . “A Change of Heart or Change of Address? The Geographic Sorting of Whites’ Attitudes towards Immigration.” Unpublished manuscript. Sears , D. O. 1993 . “Symbolic Politics: A Socio-Psychological Theory.” In Explorations in Political Psychology , edited by S. Iyengar and W. J. McGuire , 113 – 49 . Durham, NC : Duke University Press . Slothuus , Rune . 2010 . “ When Can Political Parties Lead Public Opinion? Evidence from a Natural Experiment .” Political Communication 27 ( 2 ): 158 – 77 . Google Scholar CrossRef Search ADS Soss , J. , and S. F. Schram . 2007 . “ A Public Transformed? Welfare Reform as Policy Feedback .” American Political Science Review 101 ( 1 ): 111 – 27 . Google Scholar CrossRef Search ADS Sunstein , Cass . 1996 . “ On the Expressive Function of Law .” University of Pennsylvania Law Review 144 : 2021 – 53 . Google Scholar CrossRef Search ADS Terkildsen , Nayda , and Frauke Schnell . 1997 . “ How Media Frames Move Public Opinion: An Analysis of the Women’s Movement .” Political Research Quarterly 50 ( 4 ): 879 – 900 . Google Scholar CrossRef Search ADS Trump , Donald . 2015 . “Presidential Announcement Speech.” June 16. http://time.com/3923128/donald-trump-announcement-speech/. Vasquez , Tina . 2015 . “I’ve Experienced a New Level of Racism Since Donald Trump Went After Latinos.” The Guardian, September 9. Weinberg , Jill D. , Jeremy Freese , and David McElhattan . 2014 . “ Comparing Data Characteristics and Results of an Online Factorial Survey between a Population-Based and a Crowdsource-Recruited Sample .” Sociological Science 1 : 292 – 310 . Google Scholar CrossRef Search ADS Author notes I thank Hana Brown and Emilce Santana for generously providing careful review and valuable comments. I also thank Amanda Mireles for her help securing data access. Jenefer Jedele and Peter Chu provided research assistance. All errors are uniquely my own. © The Author(s) 2018. Published by Oxford University Press on behalf of the University of North Carolina at Chapel Hill. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

Social ForcesOxford University Press

Published: Apr 27, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off