Economic Hardship or Perceived Status Threat? ROUGH DRAFT: NOT FOR CITATION OR QUOTATION; REFERENCES CURRENTLY OMITTED. Diana C.

Transcription

1 Economic Hardship or Perceived Status Threat? Explaining Voter Preferences in the 2016 Presidential Election ROUGH DRAFT: NOT FOR CITATION OR QUOTATION; REFERENCES CURRENTLY OMITTED Diana C. Mutz Samuel A. Stouffer Professor of Political Science and Communication University of Pennsylvania After each presidential election, people are naturally eager to know why one candidate won and the other lost. Because early efforts to answer this question typically come from exit poll analyses offered by journalists on deadline, they tend to be simplistic and univariate. For example, moral values were heralded as the explanation for the 2004 election outcome (Seelye 2004), though later analyses called this interpretation into question (see Mulligan 2008). Nonetheless, such explanations influence both winning and losing partisans understanding of what happened and why. The desire to explain a political outcome in concise terms is understandable; however, few instances of collective human behavior have singular causes. Thus far, the dominant narrative explaining the outcome of the 2016 presidential election has been that white working class voters rose up in opposition to being left behind economically. Those who lost jobs or experienced stagnant wages due to globalization punished the incumbent party accordingly. These blue collar workers repeatedly have been cited as the basis of Trump s victory. This explanation centers on personal economic hardship as the key explanation for Trump support. If the rust belt of the northeast and midwest brought about Trump s victory, then this must be because blue collar workers came to the conclusion that their economic interests would be better championed by Donald Trump. Interestingly, this explanation is now being embraced by liberals as well as conservatives. As Thompson (2016) summarized in a recent Atlantic article, Perhaps the clearest takeaway from the November election for many liberals is that Hillary Clinton lost because she ignored the working class. Others have suggested that racism is the main explanation for Trump s victory. Through his public statements about Hispanics and Blacks, Trump signaled acceptance of prejudicial views and attracted like-minded whites. According to this explanation, greater racial resentment also was triggered by having a black president, and this further fueled Trump s advantage (Kaufmann, cf. Kuhn 2016). The observation that Trump supporters are lower in social trust fits neatly within this explanation because racial diversity tends to go hand in hand with lower trust (see also Persily and Cohen 2016). Still others point to Trump s views of women and cite mysogyny as the basis of male support for Trump (Schaffner). 1

2 Explanations have also pointed fingers at societal trends such as decreasing trust in government and cynicism about major U.S. institutions. Indeed, anti-establishment sentiment meshed well with Trump s outspoken outsider persona. He offered a simplistic narrative in which all politics boils down to the people versus the powerful, a simplistic battle between a virtuous people and a small group of nefarious parasitic elites (Oliver forthcoming). Consistent with this populist rhetoric, Trump encouraged people to doubt the legitimacy of the political process itself, and journalists noted a relationship between low levels of trust in government and/or people in general and Trump support (Cerny 2016). To date, these claims have been made on the basis of aggregate demographic patterns that could occur for a multitude of reasons. For example, less educated people tend to have lower incomes, but they are also different on a host of other dimensions, including higher levels of racial prejudice. In addition, it is unclear which of these observed patterns reflect ongoing voter preferences rooted in party loyalty as opposed to defections from usual party alliances. In reality, no single factor is likely to account for Trump s victory. Moreover, many of these factors are intertwined; for example, prejudice enters into attitudes toward immigration and trade policies (McElwee 2016; McElwee McDaniel and Cohen). For this reason, multivariate approaches are necessary in order to understand what most powerfully influenced Trump support. At present, it seems fair to say that the 2016 presidential election suffers from a surfeit of explanations. The purpose of this study is to explore two important questions posed by Trump s victory. First, did voters in 2016 effectively politicize their personal economic interests in supporting Trump, as has been widely claimed? Second, what additional or alternative explanations for Trump s support are most likely? After reviewing evidence pertaining to several theories, I use two novel surveys to address these questions. Economic Self-interest and Voting Behavior Initial reasons for skepticism about the dominant interpretation of the 2016 election come from an extensive literature on when and how American voters politicize personal economic experiences. For many decades, scholars in political science and economics have sought empirical evidence of economically self-interested policy preferences. Although aggregate-level evidence was suggestive of a public that blamed incumbents for economic downturns and rewarded incumbents for economic gains, these relationships seldom held up at the level of individual policy preferences. For example, those who recently lost jobs were unlikely to blame government policy for their personal circumstances, and those who suffered financially under a given administration were no more likely to vote against the incumbent. Across a wide range of issues, scholars have found that citizens seldom form policy preferences on the basis of their immediate family s economic self-interest (see Sears and Funk for a review). This is not to suggest that citizens never do so, but the conditions under which personal financial self-interest is politicized appear to be very rare. 2

3 Thus far there is mixed evidence that economic distress predicts viewing Trump more favorably (see Rothwell and Diego-Roswell 2016). Even bivariate evidence thus far is not entirely consistent with the class-based explanation. For example, contrary to the left behind thesis, low levels of education, not lower incomes, predicted Trump support (Silver 2016). Further, low levels of education correspond with Trump support only among whites, even though minorities are more likely than whites to work in manufacturing. If Trump s message was attractive to working class interests, this was clearly not the only issue in play. Only by simultaneously considering a multitude of potential influences can scholars hope to understand and evaluate the underlying basis of Trump s support. The general failure within the discipline to find evidence of individual economic selfinterest has led to efforts to find intermediary links between family financial interests and candidate preferences. Studies have suggested that neither ascriptive group memberships nor self-identification with specific groups whose economic interests are at stake connect economic interests to presidential preference (cf., Conover 1985, Kinder, Adams and Gronke 1989). Instead, the political impact of groups may occur by virtue of perceptions of group fairness; people react more positively to an incumbent when they perceived that groups have experienced similar changes in economic well-being. The recent emphasis on inequality and differences between how people are faring at the top versus the bottom of the economic spectrum is likely to raise the salience of fairness in economic assessments. For this reason, it could matter to voters that they perceive the working class to be faring poorly even if those voters are not themselves working class. Likewise, the attention drawn to globalization s winners and losers may raise concerns about the fairness of economic gains even in a recovering economy. In this way, the widespread perception that those in manufacturing are faring poorly due to jobs going overseas may have political repercussions that are not tied to economic self-interest. In the context of opinions toward globalization, efforts to find evidence of self-interested preferences have likewise encountered dead ends. As with economic voting more generally, aggregate-level patterns have often appeared consistent with these predictions, but supportive evidence at the individual level has been in short supply. Opinions toward the desirability of international trade were not influenced by self-interest as represented by a respondent s occupational skill level or industry of employment (Mansfield and Mutz 2009). Although education and support for trade consistently covary, with well-educated people expressing more pro-trade views, education's effects may be less tied to skill level than to whites anxieties about involvement with outgroups such as foreigners or racial minorities (Mansfield and Mutz 2009). Some suggest that possessing greater economic knowledge would make people more self-interested in their preferences, but evidence of this is limited (c.f., Mansfield, Mutz and Brackbill 2017; Rho and Tomz (2016). 3

4 Perceived Status Threat What else might account for Trump support instead of, or in addition to, economic hardship? One possibility is perceived status threat. When people feel threatened, the experience a sense of uncertainty and lack of control, which they are eager to resolve. This desire drives them toward social and political attitudes that help to reassert a sense of certainty and control. Because conservative political views emphasize traditionalism and resistance to change, scholars suggest that threat generally leads to more conservative political views (e.g., Jost, Glaser, Kruglanski and Sulloway 2003). A threatening and uncertain world makes the status quo more attractive than change, which could be for the better or worse. Threat also encourages support for protective policies, such as law and order, defense and immigration restrictions (CITE). Although symbolic threats have been less well studied than the potential for actual harm, the same theory is assumed to apply. Conservatives and Republicans tend to be more sensitive to threat than liberals and Democrats (Hibbing, Smith and Alford 2013). Further, as Hibbing (2016) suggests, Trump speaks the language of threat-sensitive individuals. By suggesting that threats are all around us and that America is no longer great, Trump confirmed many Americans sense of vulnerability and offered a dominant leader as the solution. Several experimental studies have confirmed that threatening circumstances induce higher levels of authoritarianism as a result of a desire for greater control over one s environment. In one experiment, participants led to believe their ingroup was of lower socioeconomic status than another group increased their support for authoritarian leaders. In another study, experimental treatments emphasizing increasing crime rates and a decline in prosperity led people to feel more threatened and increased their desire to uphold the dominance of powerful groups such as racial majorities, men, and more generally those of high status. When people see the social system itself as threatened, they are also motivated to defend the status quo by increasing endorsements of negative outgroup stereotypes. As Thorisdottir (2007) summarizes, What all these studies of the relationship between threat and political attitudes have in common is a consistent finding that threat makes people more likely to adopt political attitudes that are associated (but not necessarily synonymous) with political conservatism. Interestingly, Trump would not be considered a political conservative by traditional standards of conventional sexual and religious morality. In addition, he does not encourage respect for traditional institutions of government. Nonetheless, his emphasis on certainty, order, and protection in various forms may serve people s need to manage a sense of threat (Jost et al. 2007, Thorisdottir et al. 2007). His frequent allusions to the desirability of the past highlight his support for resistance to change in the established order. So although he does not represent traditionally conservative policies per se, if, as Thorisdottir (2007) suggests, strong adherence to the status quo is the primary mechanism by which conservatism offers greater 4

5 certainty (see Thorisdottir 2007; Jost and Hunyady 2005), then he fits the mold as a potential cure for the uncertainty associated with rapid change. The idea that perceived status threat triggers a greater desire for status quo hierarchical social and political arrangements is similar to the notion that authoritarianism is not so much an individual personality trait as an underlying tendency that can be triggered by circumstances in the environment (Stenner 2005; Hetherington). When a person with this underlying predisposition perceives a normative threat, they respond with a defense of the ingroup and reassert the importance of conformity to group norms. Importantly, this is not the same as a reaction based on self-interest or pocketbook considerations related to the ingroup. Instead, it is reaction to a perceived threat to society as a whole, a belief that their way of life is threatened by change and nonconformity. A series of experiments that purposely raised the salience of threats to the moral order produced results consistent with this argument. For example, reading an article that described Americans as changing in ways that make them more similar to one another had no effects. But reading an article that suggested Americans were becoming more morally diverse triggered greater intolerance and a greater emphasis on maintain the national order. Although this study does not focus specifically on the tendency toward authoritarianism, it similarly draws on the notion that normative threat triggers status quo views. Racial Status Threat. Two distinct forms of status threat are currently prominent in the United States. First, for the first time since Europeans arrived in this country, white Americans are being told that they will soon be a minority race in the United States. This loss in terms of the raw proportion of the population is unlikely to change white Americans status as the most well-off racial group, but it may symbolically threaten whites perceived dominance over social and political priorities. Minority groups are expected to represent a majority of the US population within a few decades. Indeed, the U.S. Census Bureau has already announced that as of 2015, a majority of children under the age of 5 are non-white (Thompson 2016). Experimental studies show that making this racial shift salient results in increased status threat to whites and greater opposition to race-based policies such as affirmative action, as well as to healthcare reform, which disproportionately benefits minorities due to their lower economic status. Racial status threat also makes whites more conservative on non-racial protective issues, for example, by increasing support for defense (Craig and Richeson 2014). Further, as suggested by Tesler (2016), the election of the first black president may have exacerbated whites fears about loss of ingroup social status. Based on a series of survey experiments, Will, Feinberg and Wetts () argue that perceived threat experienced by some white Americans explains increased support for the Tea party. Both Obama s election and demographic trends threatened the racial standing of whites in the U.S.. Experiments emphasizing the declining proportion of whites in the U.S. 5

6 population increased racial resentment and support for the Tea Party. These findings suggest the possibility that an increasing perceived racial status threat has similarly benefitted Trump. Global Status Threat. In addition to racial status threat, many Americans also feel threatened by globalization, that is, the increasing number of interconnections and interdependence between the U.S. and other countries. As one recent headline put it, The Era of American Global Dominance is Over (Fuller 2016). Whether this headline is true or false in various respects remains debatable (see, e.g., Kiersz 2015), but the perception of threat is real. Instead of viewing the U.S. as a single hegemonic force, most predict a near future in which the U.S. is just one of several dominant economic and cultural forces in the world. The China threat in particular currently looms large in many Americans fears, just as Japan did several decades ago. What form of threat Americans envision is not always clear. But whether it is a security threat or an economic threat, globalization of the U.S. economy clearly makes Americans feel less in control of their own destinies. In the past the received wisdom from studies of American political behavior has been that voting ends at water s edge. In other words, outside of foreign wars, international issues such as trade and immigration were assumed to have little if any electoral importance. However, the process of globalization has gained prominence in the public eye and was central to Trump s platform. Moreover, globalization may pose cultural as well as economic threats to Americans. Although Americans are very divided on issues of trade and immigration, as a PIPA poll recently found, Most Americans feel that that workers are not benefiting from the increase in international trade and that the needs of American workers are not being adequately addressed by US policymakers. Other studies likewise confirm that Americans see trade as something that benefits people in other countries by providing jobs at the expense of jobs for Americans (Mutz and Kim 2016). Although economists may view globalization as a winwin proposition in the long run, the American public sees it as far more of a threat. In the analyses that follow, I begin by evaluating the received wisdom on Trump support, that is, that it results from a backlash due to economic hardship. I first examine the possibility that his victory was due to job loss and other economic hardships that afflicted many Americans. Next, I consider this explanation relative to several other possibilities including heightened status threat either due to the shifting racial composition of the U.S., or due to globalization. Methods To examine these potential explanations, I use two national probability samples. I first draw on a large representative sample of Americans interviewed in October 2016, shortly before the presidential election. This survey was conducted by Amerispeak, part of the National Opinion Research Center at the University of Chicago. In addition to a large sample size (n>3000), this survey included a large number of survey items measuring economic perceptions, issue attitudes and perceived status threat, making it advantageous for these 6

7 purposes. In addition, interviewing was done in both English and Spanish, and in whatever mode the respondent found most desirable (via telephone or internet). Using the NORC sampling frame ensured a representative sample, and the survey obtained a highly respectable response rate of xx. However, as with all cross-sectional surveys, analyses based on such data involve inevitable issues of proper model specification. Have all appropriate predictors been included? Because one can never answer this question definitively, I also take advantage of a smaller national survey consisting of panel data collected online. These nationally representative surveys were fielded by KnowledgePanel of GfK, Ltd., shortly before the 2012 election, and then again shortly before the 2016 election. GfK utilizes address-based sampling to recruit online survey participants, providing online access when necessary. The disadvantage of this sample are that it was conducted in English only, and had a much lower response rate. Further, because it was part of an over-time panel sample, it under-represents the youngest citizens by 2016, since they were not eligible in previous waves. On the positive side, using panel data that include survey questions asked in both pre-election surveys, it is possible to ascertain whether change over time in specific independent variables corresponds to change over time in vote preferences at the individual level. Panel data allow the strongest possible causal inferences outside of laboratory studies, so for purposes of studying elections, they are extremely valuable. In order to eliminate most model specification issues, I use fixed effects analysis in which each person in the panel in 2012 serves as their own control for As a result, all stable characteristics of the individual, such as income, education, gender and so forth are eliminated as potentially spurious causes of association. Further, to account for variables that might change over time that are not included in the model, I consistently include a dummy variable representing all other systematic change over time between 2012 and 2016, that is, changes that affect all respondents, regardless of cause. The most efficient way to explain support for Trump in particular is to control for the extent to which support for his candidacy is simply a function of ongoing support for Republican presidential candidates. In other words, those who voted for the Republican/Democratic candidate consistently in both 2012 and 2016 need no further explanation. Previous candidate preference is obviously a strong predictor of 2016 preference due to party loyalty. What is critical is those who deviated from a consistent pattern. For this reason, change in candidate party preference is the dependent variable of interest. I examine three indicators coded in the direction of support for the Republican candidate. As in the previous analysis, I use the feeling thermometer advantage for the Republican relative to the Democratic candidate in both election years. In past elections, Feeling Thermometer Advantage has been a highly reliable predictor of vote choice. Second, I use a measure of Republican Candidate Preference that draws on responses from all survey participants, asking for their preference among the presidential candidates, regardless of 7

8 whether they say they intend to vote or not. As a third dependent variable, I use self-reported Republican Vote Intention (or in some cases post-hoc reports of early voters) among those who claimed to have voted or plan to vote in November. 1 Cross-sectional Model To explain variance in these three dependent variables, I organize predictors in the cross-sectional analysis into seven categories: party identification, demographics, economic hardship, global status threat, racial status threat, anti-elitism, and geographic economic context. The first and typically the strongest predictors of vote choice are Republican or Democratic party identification. Second, I enter a collection of demographic variables, including age, religiosity, education 2, gender, and race. Third, I include a set of variables related to the economy and financial hardship, including income, whether the respondent is looking for (more) work, perceptions of whether their family finances are getting better or worse, and whether the national economy is better or worse. In addition to retrospective perceptions, two items assess issues of future financial concern. One index was comprised of three common sources of concern about being able to pay for college, concern about saving for retirement, and concern about being able to afford needed healthcare. In addition, another index tapped support for increasing the strength of the safety net to help people suffering financial hardships, and support for cutting taxes, even though this will result in reducing the strength of the safety net (reverse coded). Survey questions, variable construction and assessments of the reliability of these measures are included as Appendix A. Global Status Threat. In addition to these three groups of variables (party identification, demographics, and economic considerations), two separate groups of predictors evaluated the impact of status threat. Global Status Threat was assessed using six indicators. First, a widely used index of perceived National Superiority was included with the assumption that Americans who feel they deserve more or have a right to dominate other countries will be most threatened by rising internationalism. Another index consisting of five questions assessed support for Isolationism, that is, a desire to stay out of international affairs by avoiding humanitarian as well as other involvements. A single-item indicator asked whether the respondent viewed China as an Opportunity or as a threat to the United States. A fifth indicator assessed agreement with the general sentiment that the American Way of Life is Threatened. 1 Although self-reports are known to over-represent the percentage of the public that actually votes, based on voter validation studies, the reason for this appears to be that voters are over-represented in political surveys rather than a desire to report socially desirable responses (Krosnick). In other words, respondents are telling the truth, and surveys simply under-represent those uninterested in voting, even when their demographics are highly representative. 2 Education measures were available on a seven-point scale, but only the distinction between college-educated and non-college educated respondents mattered. For this reason, education is included in all models as a dummy variable representing lack of college education. 8

9 Respondents favorability toward immigration was measured using questions asking about support for a path to citizenship, opposition to building a wall between the U.S. and Mexico, and opposition to deporting illegal immigrants. The final indicator included within Global Status Threat was an index comprised of two questions measuring Support for International Trade. Racial Status Threat. The second block of threat indicators included well-known operationalizations of Racial Status Threat. An overt measure of intergroup bias asked respondents to direct endorse positive versus negative stereotypes pertaining to three domestic racial and ethnic ingroups and outgroups in the United States. An index of Social Dominance Orientation measured respondents preferences for unequal relationships among groups of people. As an enduring individual difference with high test-retest reliability, it is a well-established predictor of hierarchy-enhancing attitudes such as belief in the inevitability of patriarchy and white racial dominance (Fiske and North). Those high is social dominance exhibit biases in favor of high status groups. It has the advantage of being more broadly applicable than direct measures of racial prejudice, yet also avoids the risk of triggering socially desirable responses that occurs when asking directly about race. Thus the survey includes measures of both directly expressed prejudice toward racial outgroups and an index that more covertly taps belief in the status quo racial hierarchy. Because low levels of social trust have been argued to predict Trump support as well as going hand in hand with perceiving one s environment to be racially heterogeneous and threatening (Putnam), a standard Social Trust Index also was included. Two novel indicators were constructed to represent the extent to which a respondent believes that high status majority groups, such as whites and Christians, are discriminated against in favor of lower status minority groups, such as Blacks or Muslims. Respondents were asked for their perceptions of the extent of discrimination against all four of these groups. These responses were used to create two indicators of majority discrimination; one addressing perceptions of discrimination against the majority Christian religion vis a vis Muslims, and the other measuring perceptions of discrimination against majority whites vis a vis Blacks. Anti-Elitism. The sixth group of indicators evaluate anti-elitist, populist sentiment that has also been credited for propelling Trump to the presidency. A standard Index of Trust in Government serves as one such measure, but with obvious caveats. As is the case with perceptions of the national economy, scholars have shown that people s trust in government is heavily influenced by the party currently in power, just as the national economy is perceived to be stronger by in-party partisans. Thus Republicans are expected to trust government less with a Democrat in the White House. Although I evaluate the effects of trust in government above and beyond party identification, there remains considerable potential for endogeneity. In addition to Trust in Government, a second index of four questions tapped beliefs in the legitimacy of the U.S. system of government. Trump s campaign stressed a broken system lacking in legitimacy, thus those viewing the system as in need of serious changes should be more likely to support Trump. 9

10 Geographic Economic Context. Finally, objective indicators of the economic context in which a respondent lives are included in the seventh group of variables. Based on respondents zip codes, each respondent was matched to the percent civilian unemployment, percent manufacturing employment, and median income in his or her immediate area. These measures of geographic context are designed to take into account the fact that even if one is not personally experiencing economic hardship, living in a community that is economically depressed may, nonetheless, have consequences for people living there in terms of their property values, quality of life, and so forth. To the extent that context matters, it should be those in areas dominated by manufacturing, areas with low median incomes and high unemployment that are most drawn to Trump on the basis of economic hardship. In describing the seven groups of explanatory variables used in this analysis, I should note that I do not intend to suggest that these groups of variables are independent explanations. For example, support for a stronger safety net could be interpreted as support for left behind manufacturing employees, or it could be interpreted as support for minority groups if one has a racialized social program in mind. And precisely because social dominance scales do not specify the kind of group hierarchy in question just that it is appropriate that some groups dominate others it may be relevant to global status threat as well as to racial status threat. Those with high levels of social dominance orientation are known to oppose trade and foreign direct investment (Mutz and Kim 2016) as well as to express prejudicial attitudes toward domestic minorities (Mansfield and Mutz). Perceiving threat from outsiders, whether racial outgroups or international outgroups, appears to go hand in hand. Likewise, although I have included immigration opinions and trade attitudes within the category of Global Status Threat, given that both issues involve interactions with foreigners, opinions on these issues are known to be driven by racial prejudice as well (Sabet 2014; Mansfield and Mutz 2008). A tendency to think of the world in terms of us versus them, ingroup versus outgroup, transcends specific types of groups. In addition, some have suggested that immigration views are really about prospective economic threat (Gerber, Huber et al), whereas others believe such attitudes have little to do with personal financial threat (e.g., Hainmueller and Hopkins). Thus these categorizations, while helpful for purposes of conceptual organization, are not to be understood as hard and fast boundaries. Panel Model The October 2012 to October 2016 panel includes over 1200 participants who responded in both waves. In addition to tapping support for the Republican and Democratic presidential candidates running in these respective years, the surveys also include the same questions covering a variety of issue opinions, perceptions of the Republican and Democratic candidates, and economic conditions in the country as a whole. Because these repeated items provide leverage in examining whether changes in respondent s perceptions over time resulted in changes in their preferred candidate, the panel analysis focuses on items that appear in both 2012 and 2016 surveys. 10

11 Fortunately, a wealth of questions were asked in identical fashion in both waves. These include items tapping personal economic hardship, such as whether the respondent was looking for work at either point in time and income at both points in time. Subjective assessment of personal financial hardship included respondents perceptions of whether their personal economic circumstances were improving or declining, and whether they view trade as having affected their family financial situation positively, negatively, or not at all. Regardless of whether they are accurate in their perceptions of the consequences of trade for themselves, they allow an assessment of whether support for the Republican candidate in 2016 was affected by these perceptions after taking into account respondents 2012 presidential preferences. Likewise, it is possible to evaluate perceptions of change over time in respondents personal economic well-being on the likelihood that they voted Republican in Retrospective perceptions of the U.S. economy as a whole were also tapped exactly as in the cross-sectional survey. Finally, three parallel measures were available for each of four different issue areas including attitudes toward trade, immigration, China as an opportunity or threat, and support for healthcare reform. For each of these issues, respondents were first asked to locate their own position on a seven-point issue attitude scale. Subsequently, they used the same scale to identify where they perceived the Republican and Democratic candidates to be on these issues in both years. There may be a tendency to assimilate a candidate of one s own party and thus place his or her views closer to one s own, and/or to contrast the views of one s opponent and place him or her further away on the scale (King 1976, Granberg and Brent). This would render perceived distance measures endogenous to candidate preferences. Fortunately, this issue is less problematic for purposes of panel analyses. Using fixed-effects panel analysis, I compare each respondent to himself or herself at a previous point in time. Thus assuming the tendency to assimilate or contrast is constant for a given individual, any tendency to do so with a candidate s perceived issue position will occur at both points in time, thus canceling itself out in an over time, within-person analysis. Moreover, all unchanging characteristics of individuals, such as demographic and personal traits, will fall out of a fixed effects model because traits that do not vary cannot explain change over time. Given that I am attempting to explain changes in voter preferences from the candidate of one party to another, panel analyses are ideal. They also eliminate model choice issues based on potentially spurious relationships that can occur in cross-sectional analyses. In addition to issue opinions and perceptions, the panel data also included a subjective assessment of the extent to which a respondent feels that trade in particular has negatively affected his or her family financial situation, and the extent to which the respondent feels his or her family financial situation has gotten worse/better, regardless of the reasons. Likewise party 11

12 identification was assessed at both points in time on a seven point scale, as were perceptions of whether the national economy was improving or declining. 3 Results Cross-Sectional Analyses I begin by using the cross-sectional data to evaluate the extent to which Trump s support was predicted by economic hardship. As shown in Table 1, I limit predictors in this analysis to economic variables in order to maximize the chances that they might demonstrate some predictive power without competition from covariance with other independent variables. As shown in Table 1, across the three separate operationalizations of Trump support, findings were highly consistent. First, Republican and Democratic identification had their anticipated influence. Lacking a college education has the widely noted positive impact on Trump support, and being white likewise raised levels of support for Trump. Gender had its predicted negative effect on support for Trump in two out of the three analyses. On the other hand, indicators of Economic Hardship did not have their anticipated effects. When income was related to Trump support at all, it was in the direction opposite what economic hardship would predict, with high income citizens more likely to support Trump. Those looking for work were no more likely to support Trump than those gainfully employed or retired. Subjective assessments of one s own finances only demonstrated the predicted direction of influence in the case of Trump Thermometer Advantage. If economic hardship were a voter s major concern, one would expect greater support for a safety net for displaced workers. Yet across the board in Table 1, Trump support is predicted by opposition to a stronger safety net, regardless of which operationalization of the dependent variable is used. Despite this generally unsupportive pattern, two findings in Table 1 are consistent with the economic hardship thesis. First, those with greater concern about meeting future expenses for education, healthcare or retirement were consistently more supportive of Trump. In addition, when evaluating the impact of the respondent s immediate geographic context, one of the three economic indicators showed signs of the predicted relationship. Consistent with the journalistic narrative, Trump support was much greater among people residing in areas high in manufacturing employment. Although this relationship is more subject to endogeneity, perceptions of the national economy exhibited their usual relationship with support for the 3 Direct assessments of domestic racial ingroup prejudice were also available in both waves, but the large number of missing values when combining the two waves made it prohibitive to include these measures in the panel models. Further, there was no clear prediction related to change over time in prejudice given that increases in prejudice should make Republicans less likely to defect (due to their agreement with the presidential nominee), and Democrats more likely to defect (due to their greater liking for a Republican candidate who expresses views on race that are similar to their own). 12

13 incumbent party, with those who perceived the economy to be gaining ground less likely to support Trump. Tables 2, 3 and 4 illustrate the changes that occur once potential confounding influences are taken into account in analyses involving each of these outcomes. The first consistent finding worth noting is that education drops out of the equation altogether. The less educated are no more likely to support Trump once other factors are taken into account. White people on the other hand, are still more supportive of Trump. In two out of three cases, income is significantly predictive of Trump support, but as in Table 1, the coefficients suggest that higher incomes predict greater Trump support. Again, as in Table 1, opposition to a stronger safety net is a strong and consistent and predictor of Trump support, contrary to the idea that he was supported for championing those down on their luck. Shifting to the explanatory variables associated with perceived threat, the findings are more consistent with expectations. Those who perceive the American way of life to be threatened are more supportive of Trump in two out of the three analyses. Direct measures of racial prejudice approach significance in all three analyses, but the impact of ingroup favoritism is much clearer with indirect measures of social dominance orientation. In all three cases, high levels of social dominance are strong and significant predictors of Trump support. Those who believe that certain groups should dominate others are far more attracted to Trump s candidacy. Consistent with this thesis, a belief that majority whites are more discriminated against than Blacks is one of the most strongly predictive variables in the analysis. Opposition to more liberal immigration policies is likewise a strong and consistent predictor of Trump support. Are such views necessarily indicative of perceived status threat? It is possible, for example, that opposition to immigration is an economic concern. For example, agreement with the statement that immigrants will take jobs away from Americans was found to be a strong predictor of Trump support. However, the fact that those looking for work were no more likely to support Trump speaks against this interpretation. Likewise, the fact that income positively predicts Trump support calls into question the role of personal economic threat since newly arrived immigrants are unlikely to be hired into high paying jobs. Opposition to international trade is also a powerful predictor of support for Trump. In all three analyses, this index has the anticipated effect of increasing Trump support. But is this view based on the economic hardships that have befallen these voters due to trade, or to generalized dislike for interactions with foreign nations? I set this question aside for the moment given that the panel analyses are able to speak more directly to this question. But as with immigration, if the concern were tied to personal economic hardship, one would expect that those whose personal financial situations are declining, or who are looking for work to be more supportive of Trump, but such was not the case. Finally, in addition to support for the role of perceived racial and international threats, there is also some evidence that support for Trump is a form of anti-washington elitism. Those 13

14 who trust government the least were far more supportive of Trump than those with greater trust in government. Although this relationship could be a sign of his populist appeal, others have suggested that high levels of current government distrust instead reflect distrust of a black president, thus tying this indicator to perceived racial threat. Panel analyses In evaluating effect sizes in panel data, one must take into account whether change in a given independent variable predicts change in the dependent variable, as well as how much that independent variable actually changed during the course of the panel. Even large panel coefficients could indicate little net influence if the independent variables associated with them experienced next to no change over time. Thus to begin this analysis, I describe the percent of voters who voted for the candidate of the same party in both 2012 and In other words, these are people whose 2016 preferences are simply an extension of their previous preferences. If one considers only those voters who reported voting for either Obama/Romney and Clinton/Trump (that is, excluding third party voters and undecideds in both years), then the power of party loyalty appears quite strong. Ninety-two percent of those who voted for Obama in 2012 also voted for Clinton in If instead one considers consistency among those who voted for the Democratic candidate versus another candidate, then the percentages are slightly lower, with only 86 percent of those who voted for Obama in 2012 also voting for Clinton in 2016, and 89 percent of those who voted for Romney in 2012 also voting for Trump in Nonetheless, given that Trump was not a Republican establishment candidate, these numbers are impressive in what they suggest about voter consistency over time. More to the point of the current analyses, they set a very high bar for explaining any remaining variance. Given that most of the variation in 2016 candidate preference is explained by 2012 candidate preference, there is relatively little left to explain. Despite the fact that 2016 is widely thought of as a non-traditional election, ongoing party loyalty was still the most important factor. To produce even more conservative estimates of impact, the fixed effects model also allows party identification to vary over time along the full seven-point scale, so variance due to respondents who became more conservative, for any reason, is accounted for directly by changes in party identification. Overall, the panel shifted significantly in the Republican direction over the course of these four years, consistent with the hypothesis that threat leads to greater conservatism. But beyond changes in partisanship from 2012 to 2016, or changes in perceptions of the national economy, what accounts for defection in favor of Trump or against Clinton among previous Obama supporters? Using a fixed effects model of within-person change, Tables 5, 6, and 7 illustrate the extent to which change in respondents opinions, the distance of their own 14

15 opinions from the two major candidates, declining personal finances, and the personal impact of trade are responsible. Without summarizing each individual finding, these tables make three important points. First, declining perceptions of family finances had no bearing on defections to Trump. This is clear from the coefficients corresponding to the impact of personal economic well-being as well as from the coefficients attached to perceptions of the personal impact of trade. Although respondents subjective assessments of the personal impact of trade approach significance in two of these three analyses, they do not explain significant variation in Trump support. Likewise, changes in the perceived legitimacy of the democratic system do nothing to explain Trump s rise. Although conservatives tend to be less trusting, there was no shift toward less trust and the changes in system legitimacy that did occur did not drive Trump support. Although Republicans have long been more distrusting of the system, this was nothing new or influential. A second finding of note is that defections were not driven by changed opinions among the public on these issues. Of the 12 potentially influential coefficients corresponding to the impact of change in respondents opinions, only one is significant, and in a direction opposite what was predicted. Instead, the changes that drove greater support for the Republican candidate in 2016 were changes in how the Republicans and Democratic candidates were viewed relative to individual members of the public. As shown in Table 5, greater distance from the Democratic candidate consistently drove Trump support in a positive direction, whereas distance from the Republican consistently drove voters away from Trump. All of these coefficients demonstrate that change over time in the candidates perceived positions had a significant influence on Trump support. The results in Figures 6 and 7 are similar, although they suggest that the perceived distance of respondents from Trump is what was particularly influential. Additional support for the 2016 Republican nominee came especially from those who perceived him to be closer to their views than was Romney in Figure 2 illustrates average changes in issue positions and perceptions of the candidates positions from 2012 to As of 2012, the Democratic and Republican candidates were perceived to be virtually identical to one another in their positions on trade. It could not be an effective election issue because there were no perceived differences between the candidates on this issue. By 2016, Americans were significantly more negative toward trade, and they were much closer to the Republican candidate s perceived position than the Democratic candidate s position. Americans own opinions became significantly more negative in their own opinions toward trade and healthcare reform, but they did not shift significantly in their attitudes toward China. Somewhat surprisingly, immigration attitudes became slightly more supportive from 2012 to What is obvious from Figure 2 is that from 2012 to 2016, the positions of the Republican and Democratic candidates polarized by moving in opposing directions for three out of the four issues. The only issue for which perceptions of the candidates move in the same 15

16 direction is healthcare. Both the Republican and Democratic candidates were perceived to be more negative toward Obama s healthcare plan by 2016 relative to In short, the Republican candidate in 2016 successfully altered the party s perceived issue positions in a more conservative direction. Although it is not clear what conservative means exactly with respect to trade given Republicans usual free market stance, it seems that the shift toward closed borders for both products and people does correspond to a protectionist reaction to outside threat, thus fitting the predictions of the perceived threat hypothesis. On issues involving trade and international affairs, elites have for a long time been more positive than the mass public toward international involvement. But in previous election years, these issues were seldom front and center when it came to voting. Discussion This study provides both methodological and conceptual advantages in examining the many sources of Trump s support. First, in operationalizing key concepts, it includes multiple item measures of known reliability rather than single item questions of unknown reliability. In addition, it draws on a large number of variables as potential explanations for Trump support. Further, these are likely to be conservatives estimates of these variables impact because these models may over-control for change by including shifts in party identification and other potentially endogenous measures in the model, particularly in the cross-sectional analyses. In addition, including evidence based on panel analysis of change over time strengthens casual inference and demonstrates the importance of changes in the Republican and Democratic candidates distance from individual voter s views on several issues. Because the panel model eliminates the need to correctly specify an entire model in order to account for potential spuriousness, it provides a stronger than usual test of causal influence in observational data. In addition, because the model explains change over time in preferences for the Republican presidential candidate, it is an especially difficult test of the hypotheses proposed. Because of the difficulties and ethical issues involved in experimentation in the context of a presidential election, findings that address questions such as this are inevitably limited in the strength of their causal inferences. Observational data, even of the panel variety, and even when analyzed using powerful fixed models, cannot cover all eventualities. For example, while fixed-effects eliminates most concerns pertaining to potentially spurious influences, it cannot establish causal order when changes are expected to occur simultaneously. In addition, although model specification issues are much less problematic, fixed effects does not eliminate the possibility that independent variables may interact in producing change in the dependent variable (see Allison). 16

17 One consideration when evaluating the cross sectional models in particular is to what extent one suppresses the influence of variables highly correlated with other independent variables, thus distorting the impression of their importance in the final regression model. The highest bivariate correlation among these variables was.55, between census measures of the median income and the proportion of people unemployed in a respondent s immediate geographic context. Fortunately, the potential for collinearity occurred mainly within blocks of variables expected to be closely related. For example, within the Perceived Racial Threat block, social dominance orientation was highly correlated with the perception that whites were more discriminated against than blacks. And the perception of discrimination against majority whites was highly correlated with the perception that Christians are more discriminated against than are Muslims (r=.53). Likewise, with Perceived Global Threat, immigration opinions and trade attitudes were highly correlated (r=.49). With collinear independent variables, which specific items emerge as significant with these blocks may have more to do with the quality of measurement of individual concepts than with the nature of the concepts themselves. In measurement theory, concepts assessed with multiple questions are inevitably better than those that rely on the particularities of a single question because an index cancels out the unique content of individual items. Another potential concern stems from the representativeness of these data. To address this potential issue, I reran all of the analyses using population weights that correct for the demographic characteristics of those who were less likely to respond to the survey. The results indicate almost identical patterns of results in the cross-sectional analyses. In the panel analyses, weights could be used to correct demographic imperfections, but not incorporate younger people into the same. Those who were not 18 years of age in 2007 were ineligible for the first wave of data collection as well as all subsequent ones. Conclusion At the conceptual level, there are several important conclusions that can be drawn from this study. First, in a multivariate context, there is little support for the idea that those experiencing economic hardship provided the basis for Trump s support. Although Table 1 provides some evidence consistent with this claim in that greater economic concerns and a higher proportion of manufacturing jobs both positively predict Trump support, once the crosssectional model includes more than strictly economic considerations, these economic hardship indicators cease to be predictive. Further countering this interpretation in the cross-sectional data, income is positively predictive when at all related to Trump support, and respondents looking for work were no more likely to support Trump. Finally, those favoring a stronger safety net were less, rather than more, likely to be Trump supporters, countering the idea that Trump represented the champion of the down-on-his-luck common man. Even the effect of education on Trump support is rendered irrelevant once these other factors are taken into account. To the extent that education matters, it is not an indicator of financial hardship. 17

18 In the panel analyses, findings were no more supportive of the economic hardship thesis. First, those reporting a worse personal financial situation in 2016 relative to 2012 were no more likely to support Trump. In addition, those who increasingly viewed trade as having adverse effects on their family finances were no more likely to support Trump in 2016 relative to Romney in When considered in combination with the cross-sectional results, the evidence suggesting that Trump support was driven by economic hardship was thin to nonexistent. The sole significant indicator in either analysis that could possibly be interpreted as an economic effect was the impact of perceptions of the national economy as a whole. Those who perceived the economy to be improving were consistently less likely to support Trump relative to those who saw the economy as a whole as in decline. However, this well-known sociotropic influence of economic conditions does not represent the effects of personal economic hardship. Moreover, this finding is known to be susceptible to reverse causation. In addition, the overall mean change in response to the sociotropic question as well as the indicator of personal finances both changed in a more positive direction from 2012 to In other words, people were increasingly upbeat about both their personal finances and the national s economy. The extent to which respondents saw trade as having positive versus negative effects on their family finances did not change in either direction. These null findings are particularly important in that they demonstrate that lack of knowledge about international trade s personal effects is not the reason personal effects of trade are unimportant to presidential vote choice. Even when I evaluated the impact of people s subjective assessments of how they think they are affected by trade, whether correct or incorrect, those perceptions do not affect levels of Trump support. Likewise, their subjective perceptions of their own personal economic trajectories do not influence levels of Trump support. In the face of this evidence, it would be difficult to argue for an explanation based on individual economic interests. Support for the thesis that anti-elitist, anti- establishment views drove Trump support is mixed. In one of the three cross-sectional analyses, lower levels of perceived system legitimacy was associated with lower levels of Trump support. On the other hand, lower levels of government trust went hand in hand with Trump support in all three cross-sectional analyses. When analyzing using the more stringent test from the panel analyses, changing levels of belief in the legitimacy of the system of government did not predict Trump support. Indicators of government trust were unfortunately not available in the panel. Instead, the theory most consistent with these analyses is that perceived status threat drove Trump support. The impact of racial status threat is evident in the influence of multiple variables. First, the cross-sectional analyses demonstrated that domestic racial prejudice and social dominance were correlated with Trump support. In addition, those who felt that the hierarchy was being upended with whites discriminated against more than Blacks, and Christians discriminated against more than Muslims, were most likely to support Trump. The 18

19 consistent impact of white racial identification is also consistent with this theory. Although whites are more likely to support Republicans candidates regardless of election, the consistency and size of influence from race is unusual. Beyond racial status threat, perceived threat to America s global status also played a role in generating Trump support. Those who feel the American way of life is currently threatened, that America is and rightfully should be dominant on the world stage, are, not surprisingly, attempting to reassert control through tighter government controls. Just as psychologists find that experimentally-induced threat produces increased support for status quo hierarchies, perceptions of threat have made protective policies more attractive to voters. Whether those policies involve putting up walls to keep immigrants out, or tariffs to keep out foreign competition, support for Trump s model of a strong, dominant leader/savior to protect the masses and make America great again fit neatly into theories about what sells to populations that feel threatened. American insecurity whether economic or otherwise--creates a need for scapegoats. Racial minorities and foreigners are easy targets for blame, especially when many white Americans have limited interpersonal contact with them. Why does it matter whether Trump s support was driven by personal economic difficulties as opposed to a sense that one s position in the domestic or international hierarchy has suffered and must be reasserted? The economic hardship thesis is a potentially dangerous misunderstanding because if the economically beleaguered did not drive Trump support, then helping them out via job training programs targeted toward those who are victims of traderelated job loss will not satisfy them. This is not to suggest that those hurting economically do not deserve public support. But most white working class men are not victims of international trade or job competition from immigrants so much as of job elimination due to automation and the changing nature of employment in America (cite Irwin). Further, as shown in these data, Trump supporters are highly unlikely to support strengthening the safety net, despite the changing nature of the economy. Even if economic protectionism could bring back old-style manufacturing jobs (a dubious assertion), the underlying problem would remain. Any economic policies designed to help displaced manufacturing workers will disproportionately help minorities, the very segment of the population already resented by majority whites. Because minorities are more likely to be employed in these occupations, they are also more likely to benefit. Given that minorities are already seen by whites as benefitting disproportionately from redistribution policies, resentment would only increase. To the extent that they feel threatened by economic progress among outgroup members in the U.S. and by the rising middle class in other nations, they are unlikely to find solace in their improving personal financial circumstances. In many ways, a sense of societal threat is a much tougher opponent than an economic downturn because it is a psychological mindset rather than an actual event or misfortune that has befallen some citizens. Given current demographic trends within the U.S, minority influence should only increase with time, thus heightening this source of perceived threat. 19

20 Further, economists concur that international economic integration is here to stay and are only made easier by improvements in communications and transportation. Taken most literally, these findings suggest that instead of focusing on restricting trade and immigration, it might be better to address white Americans need to feel dominant over minorities and over other countries. The move toward a global economy has already left the station. Traditionally dominant countries like the U.S. may feel threatened by a rising middle class elsewhere, but this stems not from a belief that other countries are so wealthy that they do not need economic improvement so much as a sense that Americans are more deserving of jobs and prosperity than those elsewhere. In other words, a sense of American exceptionalism pervades these reactions. Interestingly, so ingrained is the competitive mindset in American judgment that when the US and other countries both benefit from trade, Americans see this as less beneficial to the U.S. than when only the U.S. benefits. In other words, the tendency is to compare one s self to another and to gauge one s relative well-being rather than absolute wellbeing. This means that it is not enough to do well; Americans have a need not just to feel they are doing well, but to feel they are doing better than others. Trump s victory is obviously viewed more admirably when it is attributed to a groundswell of support from previously ignored workers than when it is attributed to those whose already high status is threatened by minorities and foreigners. Unfortunately his victory seems more likely to validate those fears than to assuage them. The us versus them rhetoric of Trump s campaign did little to bring whites and minorities together to view one another is less theatening ways, nor did it call to whites attention the fact that they are already doing quite well relative to these particular reference groups. Instead, a message of gloom and doom reinforced the sense that it would be best to turn backwards. Both race and gender predicted preferences in a fashion suggesting that high status groups such as men and whites are reacting to a sense of threat. Those with low levels of education are obviously not part of a high status group, so doesn t this contradict the expectations of perceived status threat? As demonstrated in these analyses, once other factors are taken into account, the less educated cease to demonstrate systematically higher levels of Trump support. Less educated whites in particular are more likely to believe that high status groups such as whites and Christians are being discriminated against more than low status groups such as minorities and Muslims. They are also especially likely to believe that the American way of life is threatened, even if they don t feel their own personal finances are at risk. To the extent that education is important, this is because of its relationship with these other influences. This study also bears on the long-held belief among political scientists that political elites waltz before a blind audience when it comes to international issues (Aldrich). Public opinion on trade in particular has been assumed not to matter because politicians know they will not be held accountable on low salience issues of this kind (e.g., Guissinger ). Domestic issues were assumed to be far more important. However, issues such as trade, immigration and China clearly did matter to presidential vote choice in This could be because the 20

21 contemporary American public is simply more aware of issues tied to globalization. Alternatively, this heightened salience may be the result of Trump s choice to emphasize these particular issues. Because globalization itself is unlikely to wane, these may remain important electoral issues for the foreseeable future. Finally and most critically, these results speak to the importance of perceptions of groups in the formation of mass opinion. Political uprisings are often about downtrodden groups rising up to assert their right to better treatment and more equal life conditions. The 2016 election, in contrast, was an effort by members of a dominant racial group to assure their continued dominance, and by an already powerful country to assure its continued dominance. Perceived threats to that dominance have triggered a greater desire to uphold the status quo hierarchy, and perhaps to even turn back the clock to the status hierarchy of decades past. 21

22 Figure 1. The Role of Race and Education in Trump Support A. Probability of Support Not a college graduate Non-white White College graduate B. Feeling Thermometer Advantage: Not a college graduate Non-white White College graduate Note: Data are from October 2016 Amerispeak/NORC sample. The main effects of race and education are both significant above, as is the interaction between race and education. 22

23 Figure 2. Issue Positions of Average Voter and their Perceptions of the Republican and Democratic Candidates, (Panel data) Trade: Healthcare: China: Immigration: 23