Race and policing in the 2016 presidential election: Black lives matter, the police, and dog whistle politics

Abstract

A series of deaths of Black Americans at the hands of the police sparked mass protests, received extensive media coverage, and fueled a new civil rights movement in the years leading up to the 2016 presidential election. Both major party nominees campaigned on issues of race and policing in different ways. Drawing on colorblind racism theories and the history of law‐and‐order politics, we explore how views of race relations and the police were associated with voting behavior. We ask, on the one hand, whether people were engaged with the civil rights issues raised by Black Lives Matter and, on the other hand, whether Trump's expressions of support for the police functioned as a racial “dog whistle” to mobilize a particular set of voters. Using the 2016 American National Election Studies (ANES) Time Series Study, we find that concern about biased policing and support for the civil rights movement seeking to address it were associated with increased turnout among Democrats and more votes for Clinton. In addition, consistent with a dog whistle effect, claims of supporting the police were connected to votes for Trump mainly among those with high levels of racial resentment. We conclude by discussing the symbolic role of police in American society and politics.

Several important events during the second decade of the twenty‐first century in the United States echoed events in the 1960s (Kennedy & Schuessler, 2014; Samuels, 2014; Whack, 2017). Just as violent, frequently fatal confrontations between Black Americans and police officers sparked dozens of riots in cities across the country during the “long, hot summers” of the late 1960s (National Advisory Commission, 1968), the deaths of a series of Black Americans at the hands of police between 2014 and 2016 sparked mass public protests and once again catapulted civil rights issues to the forefront of the national agenda (Gately & Stolberg, 2015; Thorsen & Giegerich, 2014). A diverse coalition of activists joined together under the banner of Black Lives Matter (BLM) and played instrumental roles in organizing public protests and targeting political campaign events (Lowery, 2016). Supporters of the BLM movement specifically demanded police reform to prevent racially biased policing and excessive use of force, but they also raised broader issues of systemic racism and social problems faced by other marginalized groups (Black Lives Matter, n.d.; Cobbina, 2019; Lowery, 2016). Those involved in the counter‐response to BLM used traditional and social media to promote slogans including “all lives matter” to challenge the civil rights claims as well as “blue lives matter” to express solidarity with police and highlight the dangers police officers face (Bacon, 2016; Carey & McAllister, 2014; Markon, Nirappil, & Lowery, 2016). Black Lives Matter and Blue Lives Matter offered explicitly competing frames. The police were variously portrayed as heroes under attack or as perpetrators of systemic racism. BLM protestors were variously portrayed as civil rights activists or as “cop‐hating thugs.”

During the 2016 presidential campaign, Democratic nominee Hillary Clinton and Republican nominee Donald Trump similarly employed competing rhetorical frames intended to mobilize the support of voters who shared their perceptions of BLM and law enforcement (e.g., Hill & Marion, 2018). Clinton aligned herself with the BLM movement (Glanton, 2016). After the high‐profile death of Freddie Gray, who sustained fatal injuries while in police custody, Clinton gave a speech in Baltimore calling for police and criminal justice reform (Bouie, 2015; Grawert, 2016). In the first presidential debate, Clinton emphasized support for the police generally, especially all the “good, brave police officers.” She also raised a specific concern, however, that race “still determines too much,” including “how [people are] treated in the criminal justice system” and that “everyone should be respected by the law,” but she then tempered this language by emphasizing that “everyone should respect the law” and that “implicit bias is a problem for everyone” (Blake, 2016, para. 251–255). In short, Clinton attempted to strike a balanced position, acknowledging evidence of systemic racism in the criminal justice system while affirming general support and respect for police officers.

In contrast, Donald Trump was both explicitly and affectively pro‐police. In speeches, Trump frequently declared that “we love our police officers” (Nuzzi, 2016, para. 15) and “I love the police, they're the greatest” (Parker, 2016, para. 7). Trump accused Clinton, BLM protesters, and much of the political Left of being motivated by anti‐police feelings (e.g., Alcindor, 2016). He opposed BLM, saying in September 2015 that, “I think they're trouble. I think they're looking for trouble” (Campbell, 2015, para. 2) and commenting that a BLM protestor allegedly assaulted by Trump supporters “maybe … should have been roughed up” (Johnson & Jordan, 2015, para. 1). More broadly, Trump emphasized a “law‐and‐order” theme throughout his campaign by portraying crime as an out‐of‐control problem. When he officially accepted the Republican nomination, he said, “An attack on law enforcement is an attack on all Americans. I have a message to every last person threatening the peace on our streets and the safety of our police: When I take the oath of office next year, I will restore law and order to our country” (quoted in Bacon, 2016, para. 2).

In short, the 2016 presidential candidates made competing claims about race relations and policing in apparent efforts to mobilize particular groups of voters. Our interest is in how these issues were connected to actual voting behavior among the public. We use a nationally representative survey of the voting‐eligible population conducted around the 2016 election to explore this question, considering the ways that public attitudes toward the police, race, and the BLM movement were connected to voting and vote choice.

Based on the statements of the candidates—and informed by scholarly work on colorblind racism and the history of law‐and‐order politics—we focus on two separate ways these issues may have been connected to voting behavior. First, we explore whether support for the BLM movement and concern about its central issue—racially disparate policing—was associated with greater turnout among Democrats and more support for Hillary Clinton (the candidate who more closely aligned herself with the movement and its concerns). Second, we investigate the role that support for the police and racial attitudes played in generating turnout among Republicans and votes for Donald Trump. In particular, we consider the possibility that Trump's rhetoric about support for the police served as a dog whistle for voters concerned about the relative status of Black versus White Americans as media commentators, scholars, and BLM activists argued at the time (Bacon, 2016; Lee, 2016; Vega, 2016). López (2014, p. 4) defined dog whistle rhetoric as “speaking in code to a target audience.” Such rhetoric allows for politicians to speak about taboo subjects while retaining plausible deniability that they violated any social norms. In the post–Civil Rights era, it became less socially acceptable to express openly racist sentiments or stereotypes about people of color (Mendelberg, 2001). At the same time, a key component of the Republican “southern strategy” was to use law‐and‐order rhetoric to signal to potential voters who opposed the changes sought by the civil rights movement, a practice that continued in subsequent decades (e.g., Beckett & Sasson, 2004; Tonry, 2011).

The answers to these questions matter. First and most directly, they can be helpful in evaluating some of the competing public narratives about the meaning of race and policing in political context. Second, criminologists have long argued that public views of the police—the “gatekeepers” of the criminal justice system and representatives of the power of the state—matter. In particular, perceptions of the police as racially biased can diminish their legitimacy, reduce cooperation with the police and engagement in informal social control, and increase fear of crime and crime itself (e.g., Anderson, 1999; Bobo & Thompson, 2006; Drakulich, 2013; Drakulich & Crutchfield, 2013; Kirk & Matsuda, 2011; Kirk & Papachristos, 2011; Sunshine & Tyler, 2003). Through this work, we speak to the political relevance of views of the police, highlighting a different way that views of the police may matter, and we raise troubling questions about the meaning of support for the police—and the prospects for more widespread police legitimacy—when the police are used symbolically in divisively partisan and racialized ways. The findings also have implications for the viability of police reform and criminal justice reform more generally. Finally, the results shed light on some of the ways that racial civil rights movements—and the backlashes and counter‐movements to these movements—play out in public opinion and are connected to political behavior.

1 POLITICAL RELEVANCE OF THE POLICE AND RACE

The 2016 presidential election was held in the wake of substantial national discourse about race and policing. We explore two ways in which the issues of race relations and policing may have been connected to voting behavior in this election: whether concern about racially disparate policing or support for the BLM movement was connected to turnout or votes for Democrat Hillary Clinton, and whether support for police acted as a racial dog whistle associated with Republican turnout and support for Donald Trump. Two literatures are relevant to understanding these possibilities. First, collective action frames are useful for understanding how social movement actors seek to shape public opinion in ways that advance their agenda—and often provoke counter‐framing efforts. Second, conceptions of modern racism and criminological political histories illuminate the meaning of law and order political rhetoric.

1.1 Social movements, collective action framing, and mobilization

Like other civil rights movements, BLM is not a single formal organization with one leader or spokesperson but is instead a collective of loosely affiliated activists and protesters. Common themes and concerns, however, have emerged. The movement is focused specifically on racially disparate police practices but also more broadly on systemic racism as well as on problems faced by other marginalized groups (Black Lives Matter, n.d.; Cobbina, 2019; Lowery, 2016). Activists targeted campaign events beginning in 2015 with the goal of getting these issues into the conversation for the 2016 election.11 Although we are interested in the connection between Black Lives Matter and voting, we do not suggest that Black Lives Matter activists tried to influence the outcome of the election in a specific direction. Instead, their goal was to raise the substantive issues at the heart of the Black Lives Matter movement on a national political stage (e.g., Lowery, 2016). Our interest is in how these issues, once raised in such a public forum, were associated with voting behavior.

Scholars of social movements frequently focus on the ways that movement actors frame social problems to achieve specific political goals (e.g., Benford & Snow, 2000; Goffman, 1974). Framing efforts are an attempt to accomplish several core tasks: drawing attention to a particular problem, attributing specific blame for that problem, articulating a proposed solution, and motivating action around the issue (Benford & Snow, 2000). To inspire action, social movements will emphasize the severity and urgency of the problem, but they may also emphasize a moral imperative: the propriety or rightness of taking action (Benford, 1993). To this end, it can be useful to frame the harms caused by the problem as an injustice rather than as merely a misfortune (Snow & Benford, 1992; Turner, 1969). Focusing on the disproportionate likelihood of Black citizens dying at the hands of the police and highlighting cases in which fatal force was used against unarmed Black citizens are both ways of emphasizing injustice and motivating participation in the movement. Thus, BLM worked to make racial inequalities—particularly those related to policing practices—a central issue in this election, focusing on injustice frames as a call to arms on the issue.

Clinton may have seen potential political value in endorsing these views, at least cautiously. From a collective action framing perspective (Benford & Snow, 2000), BLM found a frame that resonated with a broad collection of potential voters. Clinton engaged in frame bridging tactics, linking civil rights concerns with gendered concerns that were already at the heart of her campaign—as reflected in campaign slogans like “Breaking Down Barriers,” “Fighting for Us,” and “Stronger Together” (e.g., Keith, 2016).

Framing efforts by one group typically spur counter‐framing efforts by other groups that seek to redefine in the public mind the identified problem, the cause of the problem, and the proposed solution to rally opposition. For example, to counter anti‐war protestors, military hawks often reframed the imperative to support war efforts as a matter of “support for the troops.” This counter‐framing casts anti‐war protest not as advocacy for humanitarian concerns but as disrespect for the “heroes” who protect civilians at home (e.g., Coy, Woehrle, & Maney, 2008). Opponents of BLM seemed to employ a similar tactic using slogans like “blue lives matter” to portray police officers as heroic public servants who deserve support rather than criticism.

1.2 Colorblind racism, dog whistles, and “law‐and‐order” politics

A complimentary explanation rests on scholarship about political structure, race relations and racism, social boundaries, and the politics of crime and justice. In short, the idea is that the police, in this political context, may have a symbolic meaning tied to the racial order. Specifically, it is possible that “support for the police” may function as a dog whistle specifically for individuals who are concerned about potential upsets to the existing relative position of racial groups (e.g., Blumer, 1958).

By definition, social movements advocating for racial equality present direct threats to status quo race relations. According to conflict theory, some members of groups benefitting from the existing structure of race relations will act to preserve the status quo using tools that their social and political standing make available to them—including disproportionate control over the police and judicial system (e.g., Chambliss, 1975). A perceived threat to one's group's status is critical in studies of race relations—including threats to the economic, social, or political standing of the dominant group (e.g., Blalock, 1967; Blumer, 1958). These threats find their expression in prejudice and negative stereotypes, which help justify the social exclusion of subordinate groups and the rejection of their demands for equality (Blumer, 1958).

Politicians can leverage racial threat for political gain. Race is frequently used in the political realm as a symbolic and social boundary (Lamont & Molnar, 2002). Boundaries—the symbolic lines that separate groups of people—are actively constructed to serve social or political purposes. Dividing “us” from “them” can be useful in motivating cohesion and collective action among the “us” against the threat posed by “them.” Politicians frequently seek to redraw boundaries to draw in potential supporters—the important distinctions between Democrats and Republicans have been drawn and redrawn along class, race, religion, “values,” and a multitude of other dimensions. Boundaries can also be activated to divide groups. The United States has a long history—as far back as Reconstruction—of using “racial separation” to drive a wedge between lower class White and Black Americans to prevent “a united fight for higher wage and better working conditions” (Du Bois, 1935, p. 700). More recently, the combination of a seeming decline in the social standing of Whiteness—a process begun in the Civil Rights Era and heightened first by the election of Barak Obama and then by the rise of new civil rights movements—combined with real economic harms brought on in part by global economic changes—seems to have produced a vivid sense of “ressentiment” among some White Americans (Cramer, 2016; Hochschild, 2016; Olson, 2008; Scheler, 1912/1972). This ressentiment finds its expression in an animus toward Black Americans, immigrants, Muslims, and others. As Hochschild (2016) described in her account of a deep story—an affective interpretive lens for political issues—many White Americans feel as if members of these other groups have been granted special assistance and benefits from the federal government that has allowed them to “cut the line” in which they have been patiently waiting to achieve the prosperity promised in the American dream.

Despite the political utility in activating racial boundaries, politicians today will often avoid talking directly about race. When racial inequalities are accepted, groups in power use openly racial ideologies to justify their position, but when inequalities are challenged, groups in power shift away from racial ideologies and toward those emphasizing individualism (Jackman & Muha, 1984). The Civil Rights Movement of the 1960s presented such a challenge and was successful in shifting social norms such that overtly discriminatory policies and overtly bigoted rhetoric—what Bobo and Smith (1998) described as “Jim Crow” racism—became less acceptable to the mass public (Mendelberg, 2001). The product was a “modern” form of racism, variously described as symbolic (Kinder, 1986; Kinder & Sears, 1981; Sears, 1988), laissez‐faire (Bobo, 2004; Bobo & Kluegel, 1997; Bobo & Smith 1998; Bobo, Kluegel, & Smith, 1997), or colorblind racism (Bonilla‐Silva, 2018). This modern form of racism rests on two key principles: a denial or minimization of contemporary racial discrimination and inequalities, and a focus on individualism and meritocracy that (implicitly or explicitly) blames racial disparities on people of color by arguing that they lack the proper work ethic and discipline to succeed in today's fair, equal‐opportunity society (e.g., Bobo et al., 1997; Bonilla‐Silva, 2018). Like the older, more explicit form of racism, the modern ideology motivates opposition against efforts to ameliorate racial inequalities, thereby maintaining White hegemony. Unlike the older form, those who hold colorblind or laissez‐faire racist views often do not see themselves as racist, and in fact, they may explicitly reject overt expressions of racism. Thus, individuals holding these views will strongly object to contemporary policies designed to address racial inequalities—in some cases framing these policies as racist against White Americans—while minimizing the consequences of historical policies that favored Whites (e.g., Bonilla‐Silva, 2018).

This distinction is important. Although both older and newer forms of racism motivate opposition to policies that might address racial inequalities, the newer form may not be consciously motivated by racial animus or a belief in the inherent inferiority of people of color. People who agree with the deep story Hochschild (2016) identified simultaneously eschew overt racial animus but express concerns about the racial status of White and Black Americans consistent with modern colorblind or laissez‐faire racism. Echoing Du Bois’ (1935) concerns about the wedge drawn between poor White and Black Americans, Hochschild (2016) noted that White Americans who feel that they are “falling behind” or are “strangers in their own land” tend to focus blame for their social slippage on the perceived line‐cutting of other groups and a government seen as indulging these groups with special treatment rather than, for instance, big businesses moving manufacturing abroad.

Given this social shift from overt to colorblind racism, politicians have turned to implicit appeals and the use of racial “dog whistles” to appeal to voters who experience feelings of racial threat while avoiding direct references to race (López, 2014; Mendelberg, 2001). Law‐and‐order rhetoric—including references to the police—has long been employed by politicians as a dog whistle intended to reach those concerned about threats to the racial order. This grew out of efforts by pro‐segregation politicians of the 1960s to reframe civil rights protests as promoting lawlessness, characterizing mass protests and unrest in urban areas as riots, and civil rights advocates as having no respect for law and order (Alexander, 2010; Beckett, 1997; Scheingold, 1984, 1992; Weaver, 2007). The Republican Party subsequently embraced the theme of law and order with the purpose of building a “southern strategy” intended to attract those White voters (especially southerners and suburbanites) who felt threatened by post–civil‐rights changes to the nation's racial hierarchy (Carmines & Stimson, 1989; Edsall & Edsall, 1991). Conservative politicians relied heavily on racially coded, dog whistle rhetoric to put the southern strategy into action (López, 2014), and this practice has continued in subsequent decades (e.g., Beckett & Sasson, 2004; Tonry, 2011). Presidential candidates such as Barry Goldwater, Richard Nixon, Ronald Reagan, and George H.W. Bush used the law‐and‐order theme in their campaigns with talk of being “tough on crime” and/or supporting a “war on drugs” in the “inner cities.” Critics have also argued that part of Bill Clinton's electoral success was a result of his strategic choice to be even tougher on crime and “welfare fraud” (another racialized dog whistle) than conservative Republicans (López, 2014; Murakawa, 2014). From a wealth of research, scholars have connected racial attitudes to views of the police (e.g., Barkan & Cohn, 1998; Carter & Corra, 2016; Carter, Corra, & Jenks, 2016; Matsueda & Drakulich, 2009) and other punitive attitudes (e.g., Bobo & Johnson 2004; Chiricos, Welch, & Gertz, 2004; Johnson, 2001, 2008, 2009; Peffley & Hurwitz, 2002; Wozniak, 2016), including support for the death penalty (Matsueda & Drakulich, 2009; Unnever & Cullen, 2007, 2010). As recently as the 2008 election, a strong connection remained between implicit racial antipathy and support for law‐and‐order rhetoric and policies (Drakulich, 2015a, 2015b). Bobo (2017) argued that we remain firmly in an “era of Laissez‐Faire racism” (p. S85).

The institution of policing has a complicated racial history in the United States. Police have been tasked with enforcing both racist and anti‐racist laws. On the one hand, law enforcement agents have protected civil rights protesters, enforced desegregation orders, and protected Black students as they integrated public schools. On the other hand, law enforcement officers also helped enforce the Fugitive Slave Act prior to the Civil War, facilitated the convict leasing program during Reconstruction, enforced the Black codes and Jim Crow laws during the first half of the twentieth century, and quelled mass protests for racial justice from the 1960s up through the present (e.g., Alexander, 2010; Blackmon, 2008; Drakulich & Rodriguez‐Whitney, 2018; Wacquant, 2003). As a result, the image of U.S. Marshals protecting a young Black schoolgirl in Norman Rockwell's “The Problem We All Live With” stands in contrast to photographs of officers confronting civil rights protesters in cities like Selma and Montgomery.

Unsurprisingly, then, the symbolic meaning of the police is also multifaceted. If crime represents an erosion of social norms, then people who are concerned about broader social change may “look to the police to defend a sense of order” (Jackson & Bradford, 2009, p. 499; see also Jackson & Sunshine, 2007; Wozniak, 2016). In this way, the police may be viewed as a symbol of order and lawfulness. Both crime and the police, however, are frequently racialized in American social and political discourse (Chiricos et al., 2004; Muhammad, 2010; Russell‐Brown, 2009). In this light, drawing on conflict, racial threat, and modern racism theories, it is possible that for some Americans the police might be a symbol of protection against threats to law and order that they attribute predominantly to racial minorities. At the extreme, some Americans may even see the police as defenders of a racial order.

Given these complications, it is possible for one person to admire the police for their role enforcing civil rights legislation, whereas another admires them for cracking down on “lawless” civil rights protestors. This variation and these differences in meaning are central to our interest. We draw on López's (2014) description of a dog whistle as a seemingly neutral statement that has special meaning to a subset of voters with a specific set of shared views. From this perspective, rhetoric about the police may be employed in the same way that law‐and‐order rhetoric has been historically: as a means to target and attract those voters who are concerned about the relative status of White versus Black Americans without explicitly referencing race.

1.3 Research questions

Our broad interest is in exploring the way that views of the police and race were connected to voting behavior in the 2016 election. More specifically, we are interested in the two stories developed earlier. First, whether support for the BLM movement and concern about its core issue—racially disparate policing—was associated with turnout among Democratic voters and votes for Hillary Clinton. Second, whether support for the police acted as a racist dog whistle associated with turnout among Republicans and votes for Donald Trump. Although much has been written by political scientists and penologists about the politics of law and order in twentieth‐century American history broadly, few individual‐level, empirical assessments of the relationship between public attitudes toward the police and voting behavior exist. Next, we briefly review prior research on the relationship between voter turnout, candidate choice, and the four key factors in our study (support for BLM, perceptions of police racial bias, support for the police, and modern racism), and then describe our research questions.

As a new social movement, little scholarship has directly been aimed at exploring support for BLM. A Pew survey in July 2016 found only modest support for BLM overall and among White respondents—but substantially higher support among Democrats than among Republicans. Updegrove, Cooper, Orrick, and Piquero (2018) found opposition to BLM highest among older, conservative, and Republican men, as well as among those who live in more Republican states. Thus, support or opposition to the movement seems tied to political ideology, even though its connection to voting behavior—either the decision to vote or candidate choice—has not yet been explored.

Much more research exists on perceptions of the police as racially biased, although little of this work has been focused on considering its potential political import. Criminologists have long been interested in perceptions of police injustice (Hagan & Albonetti, 1982). These perceptions are more likely among those who have directly or vicariously had negative encounters with the police (see Gau & Brunson, 2010; Hagan, Shedd, & Payne, 2005; Wortley, Macmillan, & Hagan, 1997) or been exposed to news coverage of police abuses (Weitzer & Tuch, 2004a, 2004b). On the other side, those possessing racial animus toward Blacks are less likely to view the police as acting in biased or unjust ways (e.g., Matsueda & Drakulich, 2009; Peffley & Hurwitz, 2010). Simply being exposed to information about racial disparities is unlikely to influence views of the police for many (Mullinix & Norris, 2019). Although the results of descriptive historical work indicate the police were politically relevant in elections in the 1960s, few direct tests of the role of perceptions of the police in voting behavior exist. Prior research findings reveal that contact with the police or criminal justice system may reduce institutional attachment and political participation—including voting (e.g., Brayne, 2014; Drakulich, Hagan, Johnson, & Wozniak, 2017; Manza & Uggen, 2006; Weaver & Lerman, 2010)—although fewer studies have been aimed at examining the role of perceptions of the police or the justice system. For turnout, in a study conducted during the 2016 primary, researchers found perceptions of police injustice to be positively associated with the intention to vote among liberals but not among conservatives (Drakulich et al., 2017). For candidate choice, Matsueda, Drakulich, Hagan, Krivo, and Peterson (2012) found views of the police as racially unjust were associated with a voting preference for Bill Clinton over George W. Bush in a hypothetical election (from a 2006 survey). More recently, Drakulich et al. (2017) found perceptions of police injustice to be associated with a preference for Clinton over Trump in the general election—although notably this was before either had won their party's nomination.

Although criminologists have studied a variety of views of the police—perceptions of bias, efficacy, and misbehavior, for instance—they have rarely considered affective support. In the only study we are aware of in which the political relevance of support for the police was examined, Drakulich et al. (2017) found such support unrelated to the prospective likelihood that the respondent would turn out to vote. In this same study, an affinity for the police was correlated with a prospective vote for Trump over Clinton but not significantly related once partisanship and perceptions of police injustice were accounted for. Critically, however, the potential role of support for the police as a dog whistle has not yet been investigated. The key feature of a dog whistle is that it is targeted toward a specific audience. If Trump was using pro‐police rhetoric as a dog whistle, support for the police should be related to support for Trump specifically among those with concerns about the racial order. For others, “support for the police” may indicate, in a simpler sense, actual support for the police and may not be tied to political behavior.

Finally, and relatedly, we are interested in those who have concerns about the racial order. Following the discussion from earlier, we are particularly interested in two kinds of views: racial resentment (capturing modern racism) and racial political threat. As Saggar (2007) noted, systematic studies of the relationship between racial attitudes and the propensity to vote are rare. Most of these studies have been focused on elections involving Black candidates (e.g., Krupnikov & Piston, 2015; Pasek et al., 2009; Petrow, 2010). The results of research on the 2008 election of Barack Obama, for example, indicate that racial resentment was associated with lower voter turnout (Pasek et al., 2009) and that racial stereotypes mattered among Democratic voters (Krupnikov & Piston, 2015). The 2016 election—which lacked a Black presidential candidate but included a candidate who seemed to be re‐engaging some White voters who had previously felt marginalized (e.g., Hochschild, 2016)—presents an interesting opportunity to consider the connection between racial attitudes and voter turnout. Approaching this question indirectly, Morgan and Lee (2017) reported some modest increases in White working‐class turnout in 2016, noting separately that this tends to be a group higher in racial prejudice. In short, based on this limited amount of prior work, it is possible that racial resentment and threat motivated turnout among some voters and suppressed it among others.

In addition to turnout, in prior work, scholars have frequently connected racism to vote choice (e.g., Knuckey, 2011; Knuckey & Kim, 2015; Piston, 2010; Valentino & Sears, 2005), and the results of several analyses indicate that racism may have played an important role in vote choice in the 2016 election (e.g., Abramowitz & McCoy, 2019; Drakulich et al., 2017; Fowler, Medenica, & Cohen, 2017; Hooghe & Dassonneville, 2018; McElwee & McDaniel, 2016; Sides, Tesler, & Vavreck, 2018). In fact, the deeply divided state of modern politics can be traced to the Civil Rights Movement as well as to the racist and segregationist counter‐movements of the 1960s and 1970s (e.g., McAdam & Kloos, 2014; McVeigh, Cunningham, & Farrell, 2014). Although far‐right social movements may explicitly espouse racist ideologies and agendas, more mainstream conservative political movements may deny racist motivations while embracing racist ideologies implicitly (e.g., Blee & Yates, 2015).

Our question, however, is whether racial views help explain or condition the role of views of the police. In general, racial feelings and attitudes are strongly linked to views of crime, the police, and the justice system (Bobo & Johnson, 2004; Drakulich, 2015a, 2015b; Johnson, 2001, 2008; Matsueda & Drakulich, 2009; Unnever & Cullen, 2007, 2010; Wozniak, 2016, 2018). Far‐right movements driven by anti‐minority attitudes tend to support punitive criminal justice policies (Pickett, Tope, & Bellandi, 2014; Tope, Pickett, & Chiricos, 2015). We are not aware of work, however, in which a conditional role for racial feelings in the effect of attitudes toward the police on political behavior has been directly explored.

In this study, we build on prior work on the political relevance of perceptions of the police by Matsueda et al. (2012) and Drakulich et al. (2017). Importantly, the pilot study data analyzed in the latter study were gathered in January 2016 prior to the primary elections, meaning the authors examined prospective voting behavior at a time when many voters were still considering other candidates. Drakulich et al. (2017) were also unable to capture support for the BLM movement or perceptions of Black political threat, nor did they directly explore a “dog whistle” moderating effect for support for the police—all key parts of this study's core questions.

Our general interest is in the connection between opinions about the police and race and political behavior in the 2016 presidential election. As described earlier, we are particularly interested in two potential stories about how these issues may have been connected to voting behavior. The first story is about the potential role of BLM in positively motivating political behavior—specifically turnout among those in the party whose candidate expressed support for the movement and its concerns and votes against the candidate antagonistic towards the movement:

• 1.

  Was support for BLM associated with a greater likelihood of voting among Democrats and a smaller likelihood of voting for Trump overall?

• 2.

  Was concern about police racial bias associated with a greater likelihood of voting among Democrats and a smaller likelihood of voting for Trump overall?

The second story is about the role of affective support for the police and the possibility of a racist dog whistle effect:

• 3.

  Was affective support for the police associated with voting among Republicans and voting for Trump overall?

• 4.

  Is this relationship specific to those with concerns about White hegemony: those with racial economic resentment or concerns about Black political power? In other words, do concerns about the racial order moderate support for the police?

• 5.

  Do these racial concerns also moderate support for BLM and concerns about police racial bias?

2 DATA, MEASURES, AND METHODOLOGY

2.1 Data

We analyze data from the 2016 American National Election Studies (ANES) Time Series Survey. The survey is ideal for our purposes: In addition to questions about voting behavior, it includes a series of questions about the police, race, racism, and the BLM movement. The survey was conducted in two modes: face‐to‐face interviews as well as questionnaires administered online, and in two waves—the first in the 2 months before the general election and the second in the 2 months after the election (DeBell, Amsbary, Meldener, Brock, & Maisel, 2018). The postelection interview included 1,059 face‐to‐face interviews and 2,590 online interviews. The survey is designed to be nationally representative of U.S. citizens ages 18 years or older.22 The face‐to‐face interviews were designed to be representative of the 48 contiguous states plus D.C., whereas the Internet sample was designed to be representative of all 50 states plus D.C. The minimum response rate (AAPOR RR1) was 50 percent for face‐to‐face interviews and 44 percent online. The re‐interview rate was 90 percent for face‐to‐face and 84 percent online (see DeBell, Amsbary, Meldener, Brock, & Maisel, 2018, for more information). Every state is represented by at least one respondent. Only Alaska, at 1, has fewer than 5, and only Hawaii, North Dakota, Rhode Island, and Wyoming have fewer than 10. No single state represents 10 percent of the data (California is 9.7 percent), and only Texas (7.6 percent) and Florida (5.0 percent) have more than 5 percent.

2.2 Measures

We explore two kinds of political behavior in the 2016 presidential election: whether the respondent voted, and for whom he or she voted. Voter turnout is captured simply as a 1 for those who voted and a 0 for those who did not. Vote choice was coded as a 1 for those who voted for Donald Trump and a 0 for those who voted for any other presidential candidate (less than 200 respondents voted for someone other than Hillary Clinton, and dropping these cases did not substantively change the reported findings). Less than 1.0 percent of cases were missing for voting; 2.5 percent of those who had voted were missing for vote choice. Table 1 presents means, standard deviations (SDs), and ranges for all the variables.

Table 1. Descriptive statistics

Variables	Mean%	SD	Range
Percent who voted	76%	.43	0:1
Percent voting Trumpa^a Among those who voted.	44%	.50	0:1
Percent female	52%	.50	0:1
Age	47.37	17.62	18:90
Percent married/partner	64%	.48	0:1
% separated/divorced/widowed	19%	.40	0:1
# of children in household	.66	1.09	0:9
Years of education	13.96	2.63	0:22
Income (in $1Ks)	74.57	64.48	<5: > 250
Percent unemployed	6%	.24	0:1
% evangelical/born again	22%	.42	0:1
Percent Black	11%	.31	0:1
Percent Asian	3%	.16	0:1
Percent other race	5%	.21	0:1
Percent Hispanic	12%	.32	0:1
Percent foreign born	8%	.28	0:1
Percent face to face	27%	.44	0:1
Conservative	4.16	1.46	1:7
Republican	3.80	2.15	1:7
Warm toward police	74.61	23.61	0:100
Police bias	5.33	1.35	1:7
Warm toward BLM	49.26	32.60	0:100
Racial resentment	3.19	1.13	1:5
Black political threat	1.67	.64	1:3

• Notes: Sample‐weighted for nonmissing cases. Means for dichotomous measures presented as percentages (N = 3,649).
• ^a Among those who voted.

Three questions were designed to capture different dimensions of views of the police. The first captures affective feelings toward the police, employing a “thermometer scale” question. Respondents were asked to rate their feelings toward a list of individual persons and groups on a scale ranging from 0 to 100, with 0 representing very cold and unfavorable feelings and 100 representing very warm or favorable feelings. Respondents were asked specifically how warmly or coldly they felt toward “the police.” A second question using the same format asked about feelings toward BLM. The third question taps into perceptions of police bias, asking respondents, on a seven‐item scale, whether they believe that “in general, the police treat Whites better than Blacks, treat Blacks better than Whites, or treat them both the same.” Higher values indicate the belief that police treat Whites better than Blacks. Two percent of cases are missing answers about police bias, 1.6 percent are missing feelings toward BLM, and less than 1 percent are missing feelings about the police.

We include two measures of racial attitudes. The first captures racial resentment, a dimension of “symbolic racism” widely used in prior work (Henry & Sears, 2002; Kinder, 1986; Kinder & Sears, 1981; McConahay, 1986; Sears, 1988). These views are primarily driven by social concerns about relative racial group positions (Simmons & Bobo, 2019) and are connected to both explicit and implicit indicators of racial animus (Drakulich, 2015a). They are also independent from a more general political conservatism (Tarman & Sears, 2005), which we also control for here. Scholars have used this measure of modern racism to explain views of the police (e.g., Carter & Corra, 2016; Carter et al., 2016; Matsueda & Drakulich, 2009), punitive attitudes (e.g., Bobo & Johnson, 2004; Johnson, 2008, 2009; Matsueda & Drakulich, 2009; Unnever & Cullen, 2007, 2010), and voting behavior (e.g., Abramowitz & McCoy, 2019; Drakulich et al., 2017; Hooghe & Dassonneville, 2018; Knuckey, 2011; Knuckey & Kim, 2015; Pasek et al., 2009; Valentino & Sears, 2005). We operationalize this variable as the average of nonmissing responses to the following four questions: whether Blacks should overcome prejudice and work their way up without “special favors,” whether slavery and discrimination created conditions that remained significant barriers for lower class Blacks, whether Blacks had gotten less than they deserved, and whether inequalities would be solved if Blacks tried harder. The second and third questions were reverse‐coded such that high values of the measure reflect greater racial resentment: the rejection of structural explanations for racial inequalities, the embrace of individualistic explanations, and a resentment of perceived line‐cutting (e.g., Bobo et al., 1997; Bonilla‐Silva, 2018; Henry & Sears, 2002; Hochschild, 2016). The second measure captures perceptions of Black political threat, using a single question that asked whether respondents believe Blacks have too little, just about the right amount, or too much influence in U.S. politics. Missing data were again rare: 2.3 percent of cases were missing for perceptions of Black political power, whereas less than 1 percent were missing for racial resentment.

We control for two measures of political beliefs: identification as more liberal or conservative and identification as more Democrat or Republican, each on seven‐point scales. The political identification measure contained 1.4 percent cases missing, whereas less than 1 percent of party identification cases were missing.

We also control for a variety of sociodemographic characteristics that are associated with political behavior: gender, age, marital status, parenthood, education, income, employment status, religion (specifically identification as evangelical or Born‐Again Christian), race‐ethnicity, and foreign‐born status. We also include a control identifying respondents who participated in the face‐to‐face rather than online version of the survey. Income was missing the most (4.2 percent) followed by age (2.6 percent) and evangelical (1.1 percent). All other controls were missing in less than 1 percent of cases.

2.3 Methodology

The ANES comprised a stratified, clustered address‐based sampling design for the face‐to‐face surveys and a somewhat simpler address‐based sampling design for the online survey. The survey included weights to account for the sample design and attrition. To account for the survey weights within the stratified, clustered sampling scheme, we ran survey‐weighted generalized linear models run using the “survey” package in R (Lumley, 2014; R Core Team, 2016).

A small amount of data was missing within the survey—as described earlier, only income was missing for more than 2.6 percent of cases. Despite the small amount of missing data, to be conservative we employed a multiple imputation strategy (Allison, 2002), which does not depend on the assumption that data are missing completely at random, rather that the data are missing at random after controlling for other variables in the analysis. To this end, five data sets were imputed in a process that involved using all the individual‐level variables from the analyses as well as several auxiliary variables to add information and increase efficiency. This includes information on home ownership, stock market investment, and length of time in the neighborhood. Substantively identical results were produced by listwise‐deleted analyses. For simplicity, results from the exploration of the interactions are presented using the listwise‐deleted model, even though results from the multiply imputed data are substantively identical.

Although the survey was conducted in two waves, the key police questions were only asked in the second wave, making the analyses effectively cross‐sectional. Even though we are interested in the question of whether views of the police and BLM influenced vote choice, it is possible the reverse is true: that these views were shaped by vote choice or, at minimum, by partisan or ideological loyalty. As such, it is particularly important to control for the respondent's partisanship and political ideology in models predicting vote choice. Additionally, concerned that people's views of the police or BLM may have been changed by a local police shooting occurring between the election and their completion of the survey, we used The Washington Post police shootings database to identify the 160 fatal police shootings occurring in this 2‐month period. We conducted a sensitivity test looking for differences between respondents living in a state where such a shooting occurred and those where one did not and found none. Our interpretation is that many people's views of the police and BLM may already have been “adjusted” by the large volume of national media coverage of police shootings beginning in 2014 and peaking in the summer of 2016.

We separately explore the decision to vote among the full population, and the choice of a candidate among those who did vote, introducing the possibility of sample selection bias for the second model. In separate analyses, we employ a two‐step Heckman sample‐selection model using “attention to politics” as an exclusion restriction33 Notably, this is an imperfect instrument in theory: Respondents may be paying more attention to politics because of their interest in the candidates. It is, however, correlated with the decision to vote but not with the vote choice. We also considered state‐level measures of the closeness of the election, which were better instruments in theory but in practice were related to vote choice (Trump was more successful in close states) but not turnout. (essentially an instrument that is associated with the decision to vote but not the choice of candidate). The results were substantively consistent with those reported here.

Finally, several of our key hypotheses—including the dog whistle effect—involve conditional relationships. As Allison (1999) and others (e.g., Ai & Norton, 2003; Breen & Karlson, 2013; Long & Mustillo, 2018; Mood, 2010; Williams, 2009) noted, comparing coefficients and interpreting interactions is not as straightforward in logit models as it is in simple linear models. Most importantly, the sign, value, and significance of the interaction coefficient may all be misleading. Although a variety of approaches have been proposed (see, for example, Table 6 in Mood, 2010), much of the work has been focused on the use of marginal effects to interpret interactions in these models (Buis, 2010; Karaca‐Mandic, Norton, & Dowd, 2012; Long & Mustillo, 2018; Mize, 2019; Norton, Wang, & Ai, 2004; Williams, 2012). Accordingly, we adapt a two‐step strategy based on the approaches described by Long and Mustillo (2018) and Mize (2019), and using R's “margins” (Leeper, 2018) and “prediction” (Leeper, 2019) packages. The first step is exploratory. We graph predicted probabilities under a variety of different values for the independent variables to get a sense of the substantive story.44 In addition to exploring different values of the variables involved in the interactions, we explore the predicted probabilities holding the other covariates at their means, holding other covariates at a standard deviation below or above their means, as well as using the actual values of the data. The figures present predicted probabilities with the other covariates held at their means, but we discuss the results of other models when they diverge substantively (in most cases the general story—for example, a positive effect among Democrats and a negative effect among Republicans—remained the same even as the specific values of predicted probabilities changed across the models). We also plot and examine the average marginal effect of each variable in the interaction across values of the other. Based on these explorations and guided by our substantive hypotheses, in the second step we then conduct more direct comparisons by examining first and second differences in the average marginal effects of one of the interaction terms across the other (and then reverse this to explore the other term). We present a graph of predicted probabilities and a table with selected first and second differences, and we discuss findings from the exploratory work where relevant.

3 RESULTS

3.1 Vote turnout

Table 2 presents the results from a logistic regression predicting whether the respondent voted in the election. Both feeling warmly toward the police and believing the police to be racially biased were associated with a higher likelihood of voting, even after controlling for the respondent's sociodemographics, politics, and perceptions of racial threat. For feelings toward the police, the association seems sizeable: For a 1 standard deviation change in feelings toward the police (equivalent to rating the police 24 points higher on the hundred‐point scale), the odds of voting increased by 28 percent.55 To facilitate interpretation, the table provides two different kinds of odds ratios as well as the average marginal effect (AME). For the dichotomous variables (gender, marriage, employment, religion, race, foreign born, and face to face), the odds ratio represents the difference between the two categories, and the AME represents the discrete change between these categories. For nondichotomous variables, the odds ratio reflects a 1 standard deviation change in the predictor and the AME reflects the average marginal effect for a 1‐unit change in the predictor. Feelings toward BLM were not associated with the likelihood of voting overall. Consistent with prior elections (e.g., Blais, 2007), older respondents were much more likely to vote, as were those with more education and higher incomes. Interestingly, higher levels of racial resentment are associated with a lower probability of voting—a finding consistent with the 2008 election (Pasek et al., 2009).

Table 2. Coefficients from models predicting voting

Predictors	b	SE	OR	AME
Intercept	–3.59***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.64
Female	.23*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.11	1.26	3.47
Agea^a Nondichotomous.	.03***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00	1.78	.50
Married/partner	.16	.13	1.17	2.44
Separated/divorced/widowed	–.43**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.16	.65	–6.66
# of children in householda^a Nondichotomous.	–.10*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.05	.90	–1.46
Educationa^a Nondichotomous.	.16***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.03	1.54	2.51
Income (in $1Ks)a^a Nondichotomous.	.01***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00	1.45	.09
Unemployed	–.39*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.19	.68	–5.87
Evangelical/born again	.04	.13	1.04	.57
Black	.48*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.23	1.62	7.37
Asian	–.52	.33	.60	–7.90
Other race	–.45	.25	.64	–6.80
Hispanic	–.13	.18	.87	–2.05
Foreign born	–.34	.19	.71	–5.21
Face to face	.14	.12	1.15	2.05
Conservativea^a Nondichotomous.	.03	.05	1.05	.51
Republicana^a Nondichotomous.	–.02	.04	.97	–.21
Warm toward policea^a Nondichotomous.	.01***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00	1.28	.17
Police biasa^a Nondichotomous.	.10*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.05	1.14	1.51
Warm toward BLMa^a Nondichotomous.	.00	.00	.95	–.03
Racial resentmenta^a Nondichotomous.	–.18*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.07	.82	–.62
Black political threata^a Nondichotomous.	–.08	.11	.95	–.22

• Note: Odds ratios (ORs) for nondichotomous predictors reflect a 1 standard deviation change in the predictor. Average marginal effects (AMEs) for dichotomous predictors reflect discrete difference between zero and one. Average marginal effects multiplied by 10² (N = 3,649). SE = standard error.
• ^a Nondichotomous.
• ^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).

Given the highly partisan nature of the election and the polarizing debate about the police and BLM, however, the effects of these attitudes on turnout may have depended on people's political party. This seems to be the case based on an analysis of interactions between each of the three policing measures and political party identification. To help us explore these interactions, figure 1 presents predicted probabilities for strong Democrats versus strong Republicans (the modal categories), whereas table 3 presents comparisons of average marginal effects for key values.66 Appendix A presents the full results from this interaction model, including interaction coefficients and significance levels, although as discussed in the methods section, these may be misleading. In the figures, the gray band represents the 95 percent confidence interval. Of the three, perceptions of police bias do not seem to depend on party identification.77 For perceptions of police bias, figure 1 (as well as other exploratory visualizations) reveals only modest differences in the effect of such perceptions between even strong Democrats and Republicans. Table 3 provides confirmation for this visual story. Perceptions of police bias have a consistently positive effect that seems slightly larger for Democrats than for Republicans (and is significant for Democrats but not for Republicans); however, these differences themselves are not significant. The top panel of figure 1 shows that feelings toward the police had almost no effect on Democrats’ likelihood of voting, whereas feeling warmly toward the police was associated with a strong increase in the likelihood of voting among Republicans. The first two columns of table 3—which present the average marginal effect of warmth toward the police across party identification—back up this story. Among strong Democrats, the average marginal effect of warmth toward the police was not significantly different from zero, but it was significantly different from the average marginal effect among those identifying as either Independents or Republicans. There is a small but significant effect of warmth toward the police among those who identified less strongly as Democrats, but these effects were still significantly different than those for Republicans.

Figure 1

Predicted probability of voting from interactions of police views and party identification (N = 3,233)

Table 3. Representative marginal effects (and differences) for interactions predicting voting

		Average Marginal Effects and Contrasts
	Party Identification	Warmth for Police	Contrastsa^a Reports which average marginal effects of warmth toward the police, perceptions of police bias, and warmth toward BLM are significantly (p < .05, two‐tailed) different across party identification (second differences). Average marginal effects multiplied by 10^2 (N = 3,233).	Police bias	Contrastsa^a Reports which average marginal effects of warmth toward the police, perceptions of police bias, and warmth toward BLM are significantly (p < .05, two‐tailed) different across party identification (second differences). Average marginal effects multiplied by 10^2 (N = 3,233).	Warmth for BLM	Contrastsa^a Reports which average marginal effects of warmth toward the police, perceptions of police bias, and warmth toward BLM are significantly (p < .05, two‐tailed) different across party identification (second differences). Average marginal effects multiplied by 10^2 (N = 3,233).
a	Strong Democrat	.04	d, e, f, g	2.57**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).		.18***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	c, d, e, f, g
b	Not very strong	.08*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	e, f, g	2.19**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).		.12**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	d, e, f, g
c	Independent‐Democrat	.12***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	f, g	1.80**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).		.06	a, e, f, g
d	Independent	.16***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	a	1.38*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).		–.01	a, b, f, g
e	Independent‐Republican	.21***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	a, b	.95		–.07*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	a, b, c, g
f	Not very strong	.25***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	a, b, c	.52		–.13**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	a, b, c, d
g	Strong Republican	.29***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	a, b, c	.09		–.19***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	a, b, c, d, e

• ^a Reports which average marginal effects of warmth toward the police, perceptions of police bias, and warmth toward BLM are significantly (p < .05, two‐tailed) different across party identification (second differences). Average marginal effects multiplied by 10² (N = 3,233).
• ^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).

The bottom panel of figure 1 shows that feeling warmly toward the BLM movement was associated with an increased likelihood of voting among strong Democrats, whereas feeling coldly toward BLM was associated with increased voting among strong Republicans. The last two columns of table 3 confirm this story: The effect of BLM is generally positive and significant for Democrats and negative and significant for Republicans, while having less of an effect among independents—indeed, the effects among Democrats, Independents, and Republicans all seem to be significantly different from one another.

Given the differences between the candidates on racial issues, we also explored whether the impact of racial resentment and threat depend on party identification. As figure 2 demonstrates, the effect of threat seems similar for strong Democrats versus strong Republicans, whereas the effect of racial resentment seems to depend in important ways on partisan identification. As the first two columns in table 4 show, racial resentment seemed to have a strong negative association with the likelihood of voting among Democrats and a strong positive effect among Republicans—effects that are significantly different from one another.

Figure 2

Predicted probability of voting from interactions of police views and party identification (N = 3,233)

Table 4. Representative marginal effects (and differences) for interactions predicting voting

		Average Marginal Effects and Contrasts
	Party Identification	Racial resentment	Contrastsa^a Reports which average marginal effects of racial resentment and black political threat are significantly (p < .05, two‐tailed) different across party identification (second differences). Average marginal effects multiplied by 10^2 (N = 3,233).	Black Political Threat	Contrastsa^a Reports which average marginal effects of racial resentment and black political threat are significantly (p < .05, two‐tailed) different across party identification (second differences). Average marginal effects multiplied by 10^2 (N = 3,233).
a	Strong Democrat	–7.14***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	c, d, e, f, g	–1.26
b	Not very strong	–5.25***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	d, e, f, g	–1.34
c	Independent‐Democrat	–3.24***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	a, e, f, g	–1.42
d	Independent	–1.11	a, b, f, g	–1.51
e	Independent‐Republican	1.14	a, b, c, g	–1.59
f	Not very strong	3.48**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	a, b, c, d	–1.67
g	Strong Republican	5.88***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	a, b, c, d, e	–1.74

• ^a Reports which average marginal effects of racial resentment and black political threat are significantly (p < .05, two‐tailed) different across party identification (second differences). Average marginal effects multiplied by 10² (N = 3,233).
• ^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).

In short, warmth toward the police was associated with an increased likelihood of voting among Independents and especially among Republicans. Feelings toward the BLM social movement had opposite effects for Democrats and Republicans, with affection driving Democratic turnout and animus driving Republican turnout. Racial resentment similarly depended on party, suppressing turnout for Democrats while increasing it among Republicans.

3.2 Vote choice

Table 5 presents predicted probabilities from the bivariate associations between attitudes toward the police, feelings toward BLM, racial attitudes, and vote choice.88 Each of the five measures is significantly associated with vote choice at the p < .001 level in complex survey‐weighted logistic regressions. To illustrate the relationship, predicted probabilities were computed for values 1 standard deviation below and above the predictor. In each case, the views were significantly associated with respondents’ likelihood of voting for Trump. Those who felt warmly toward the police, saw the police as unbiased, and felt coldly toward BLM were all substantially more likely to vote for Trump than were people who expressed the opposite feelings. The difference was particularly stark for feelings about BLM. Those who felt coldly toward BLM—rating them a 16 out of 100 on the scale from cold to warm (a standard deviation below the mean)—had a 78 percent predicted probability of voting for Trump. Those who rated the BLM warmly, by contrast, only had approximately a 12 percent likelihood of voting for Trump. Racial resentment and perceived Black political threat were also strongly related to vote choice. Those with high racial resentment and those who believed that Blacks have too much influence over politics both had very high predicted probabilities of voting for Trump.

Table 5. Predicted probability of voting for Trump

Variable	%
Feelings Toward the Police
Mixed (53 out of 100)	21
Very warm (97 out of 100)	64
Perceptions of police bias
Police treat Whites better	18
Police treat Blacks and Whites equally	71
Feelings Toward BLM
Very cold (16 out of 100)	78
Warm (80 out of 100)	12
Racial Resentment
Low resentment	12
High resentment	78
Perceived Black Political Influence
Too little influence	20
Too much influence	71

• Note: Predicted probabilities shown for 1 standard deviation above and below the mean (N = 2,862).

Clearly, however, each of these views of the police and race are related to each other and to a variety of other factors relevant to voting.99 Support for the police is positively and significantly correlated racial resentment (r = .33, p < .001) and perceptions of Black political threat (r = .18, p < .001). Perceptions of the police as biased against Blacks is also associated with racial resentment (r = –.53, p < .001) and Black political threat (r = –.47, p < .001). Warmth toward Black Lives Matter is also associated with racial resentment (r = –.59, p < .001) and Black political threat (r = –.40, p < .001). The question is whether they remain related to vote choice after controlling for one another as well as for the respondent's sociodemographics and politics. Table 6 presents results from a model predicting vote choice that included all independent and control variables.

Table 6. Coefficients from models predicting voting for Trump

Predictors	b	SE	OR	AME
Intercept	–5.04***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	1.00
Female	.18	.18	1.20	1.38
Agea^a Nondichotomous.	.01	.01	1.10	.05
Married/partner	.25	.24	1.29	1.98
Separated/divorced/widowed	.17	.28	1.19	1.25
# of children in householda^a Nondichotomous.	–.05	.10	.95	–.45
Education^†	–.08*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.04	.82	–.61
Income (in $1Ks)^†	–.00	.00	.85	–.02
Unemployed	.16	.40	1.18	1.12
Evangelical/born again	.45*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.21	1.57	3.41
Black	–1.07**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.42	.34	–8.23
Asian	–.50	.57	.61	–4.29
Other race	–.75	.39	.47	–5.88
Hispanic	–.57*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.28	.57	–‐4.35
Foreign born	–.46	.30	.63	–3.33
Face to face	.05	.23	1.05	.45
Conservativea^a Nondichotomous.	.41***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.09	1.92	3.21
Republicana^a Nondichotomous.	.69***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.05	4.79	5.44
Warm toward policea^a Nondichotomous.	.01	.00	1.16	.06
Police biasa^a Nondichotomous.	–.26**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.08	.71	–1.96
Warm toward BLMa^a Nondichotomous.	–.02***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00	.57	–.14
Racial resentmenta^a Nondichotomous.	.61***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.11	2.05	4.73
Black political threata^a Nondichotomous.	.34*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.14	1.24	2.68

• Note: Odds ratios (ORs) for nondichotomous predictors reflect a 1 standard deviation change in the predictor. Average marginal effects (AMEs) for dichotomous predictors reflect discrete difference between zero and one. Average marginal effects multiplied by 10² (N = 2,862). SE = standard error.
• ^a Nondichotomous.
• ^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).

Identification as a Republican and as conservative were unsurprisingly among the largest effects, with party being by far the strongest effect in the model: A 1 standard deviation increase in Republican identification increased the odds of voting for Trump by a factor of nearly 5. Among the other control variables, Black, Hispanic, and more educated voters were all less likely to vote for Trump, whereas evangelical and born‐again voters were more likely.

In contrast to the bivariate associations, controlling for other factors relevant to vote choice rendered the relationship between feelings toward the police and vote choice statistically insignificant. In exploratory analyses, we discovered that party identification and racial resentment were the primary factors that eliminated the independent relationship between support for the police and vote choice (when either is omitted from the model, support for the police remains significantly associated with vote choice). In other words, support for the police, in and of itself, does not seem to have been an important motivation for voting for Trump. Those who said they supported the police were more likely to vote for Trump, but this was because they also tended to be people who identified as Republican and felt racial resentment.

Perceptions of the police as biased and support for BLM, on the other hand, remained strongly and negatively associated with a vote for Trump. A 1 standard deviation increase in warm feelings toward BLM halved the odds of voting for Trump. Racial resentment and perceptions of Black political threat both significantly increased the likelihood of voting for Trump. A 1 standard deviation increase in racial resentment more than doubled the odds of voting for Trump.

The evidence so far indicates that “support for the police” may simply be a proxy for party politics and racial resentment in the context of 2016 vote choice. Given that dog whistles are targeted toward particular audiences (López, 2014), however, we also tested whether the relationship between support for the police and vote choice depended on a respondent's racial attitudes (i.e., a moderated relationship).1010 Appendix A presents the full results from this interaction model. Although we did not hypothesize them, we also explored interactions between views of the police and conservative and Republican Party identification, but none were significant. The top left panel of figure 3 reveals that feelings of warmth toward the police had little effect on vote choice among respondents with low levels of racial resentment (those 1 standard deviation below the mean in resentment). In contrast, there was a strong, positive relationship between feelings of warmth toward the police and the probability of voting for Trump among respondents with high levels of racial resentment (those 1 standard deviation above the mean in resentment). The first two columns in the top half of table 7 show confirmation for this story: For those with low or average levels of racial resentment, the effect of warmth toward the police on vote choice is not significantly different from zero.1111 Unlike the other measures used in interactions, racial resentment is captured as the average response to four questions (two reverse‐coded). Given this, and to ease interpretation, we present results for the mean as well as 1 standard deviation above and below the mean. For those with high resentment, the average marginal effect of warmth toward the police on voting for Trump was positive and significant, and this effect was significantly different from the average marginal effect of those with low or average levels of resentment.1212 Small within‐group Ns make a formal test of differences in this dog whistle effect across party identification difficult, but the results of exploratory work indicate the dog whistle had the largest effect among those identifying as Independent (in other words, those least constrained by their party identification in their vote choice). The effect of warmth toward the police did not seem to depend on perceptions of Black political power (note the parallel lines in the top right panel of figure 3 and the lack of significant contrasts in the first two columns in the bottom half of table 7).

Figure 3

Predicted Probability of voting for trump from interactions of police views and racial views (N = 2,501)

Table 7. Representative marginal effects (and differences) for interactions predicting voting for Trump

		Average Marginal Effects and Contrasts
	Variable	Warmth for Police	Contrastsa^a Reports which average marginal effects of warmth toward the police, perceptions of police bias, and warmth toward BLM are significantly (p < .05, two‐tailed) different across racial resentment and black political influence (second differences). Average marginal effects multiplied by 10^2 (N = 2,501).	Police Bias	Contrastsa^a Reports which average marginal effects of warmth toward the police, perceptions of police bias, and warmth toward BLM are significantly (p < .05, two‐tailed) different across racial resentment and black political influence (second differences). Average marginal effects multiplied by 10^2 (N = 2,501).	Warmth for BLM	Contrastsa^a Reports which average marginal effects of warmth toward the police, perceptions of police bias, and warmth toward BLM are significantly (p < .05, two‐tailed) different across racial resentment and black political influence (second differences). Average marginal effects multiplied by 10^2 (N = 2,501).
	Racial resentment:
a	Low (–1 SD)	–.10	c	–2.19*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).		–.23***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	c
b	Average	.03	c	–2.13***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).		–.16***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).
c	High (+1 SD)	.16***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	a, b	–2.13**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).		–.09**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	a
	Black political influence:
a	Too little influence	.07		–.94	c	–.10**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).
b	Just about right	.06		–2.42***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).		–.13***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).
c	Too much influence	.04		–3.62***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	a	–.16***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).

• ^a Reports which average marginal effects of warmth toward the police, perceptions of police bias, and warmth toward BLM are significantly (p < .05, two‐tailed) different across racial resentment and black political influence (second differences). Average marginal effects multiplied by 10² (N = 2,501).
• ^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).

We also considered whether attitudes toward BLM and perceptions of police racial bias were similarly dependent on racial views. The effect of perceptions of the police as biased depended on perceived Black political threat.1313 But not racial resentment, as evidenced by the relatively parallel lines in the middle‐left panel of figure 3 and the lack of significant contrasts in the second set of columns in the top half of table 7. There was little‐to‐no effect of perceptions of police bias on vote choice among respondents who believe that Blacks do not have enough political influence in society (the average marginal effect of perceptions of police bias among this group is not significantly different from zero). For those who believed that Blacks have too much political influence, however, views of the police as fair or even biased against Whites were associated with a much higher likelihood of voting for Trump (see the middle right panel of figure 3). Specifically, the average marginal effect of perceptions of police bias is significantly more negative among those who believe Black Americans have too much versus too little influence (see the contrasts in the middle columns of the bottom of table 7).

Finally, feelings of warmth toward BLM were more strongly negatively related to casting a vote for Trump among respondents low in racial resentment than among respondents high in racial resentment (see the bottom left panel of figure 3).1414 Although more of those who were highly racially resentful felt coldly toward Black Lives Matter (as reflected in the smaller confidence interval in figure 3), ∼3.7 percent of the sample were both highly racially resentful (a 4 or higher) and felt warmly toward Black Lives Matter (a 60 or higher). The effect of warmth toward Black Lives Matter does not seem to depend on perceptions of Black political power, as seen in the overlapping estimates in the bottom right panel of figure 3 and the lack of significant contrasts in the bottom right column of table 7. The top right columns of table 7 shows that the effect of warmth for BLM is consistently significantly negative but also significantly more strongly negative among those with low versus high racial resentment.

3.3 Discussion and conclusion

The 2016 presidential campaign followed several years of mass protests sparked by widespread attention to a series of deaths of Black Americans at the hands of police officers. BLM activists called for police reforms and broader societal changes to address racial inequalities (Cobbina, 2019; Lowery, 2016). At the same time, other Americans—harmed by broader economic changes and resentful at what they perceived as others cutting in line for the American dream—reacted to the BLM movement with shows of support for police officers (Hochschild, 2016, p. 289). Within this polarized climate, presidential candidates Hillary Clinton and Donald Trump took differing positions. Clinton expressed support for BLM and vowed to advance criminal justice reform to address systemic racism. Trump expressed support for the police, accused BLM and their supporters of being animated primarily by hatred for law enforcement, and vowed to restore “law and order” to America's streets. Assessed in historical context, Trump seemed to resurrect racialized themes that figured prominently in electoral politics from the 1960s through the 1990s.

Our basic question was how people's opinions about these two related political issues—race relations and policing—were connected to voting behavior in the 2016 election. Our results indicated that they were connected but in different ways reflecting the opposing sides in this framing battle. Even controlling for strong correlates of voting behavior, such as political partisanship and ideology, age, education, and evangelical affiliation, we found that respondents’ feelings toward the police, perception of police bias, feelings toward BLM, and racial resentment were significantly related to their likelihood of turning out to vote—sometimes in opposite ways for those on different ends of the ideological spectrum. We also found that the latter three factors were significantly related to respondents’ vote choice. Specifically, the results reveal two distinct sets of stories.

On the one side, Democrats who were supportive of the BLM movement were more likely to vote. In addition, those who supported BLM and shared with protesters the belief that racially disparate policing practices are a problem were substantially less likely to vote for Donald Trump, even after controlling for ideology and partisanship. In contrast to public accusations, Democratic voters did not seem motivated by a dislike or lack of support for the police.

On the other side, support for the police was associated with greater turnout among Republicans, as was opposition to the BLM movement. After controlling for political partisanship and racial resentment, we found that feelings of warmth toward the police were not significantly related to vote choice, which indicates that expressing support for the police as a motivation for one's choice to vote for Trump may simply be a proxy for partisanship and concerns about threats to the racial status quo. Even more interestingly, we found evidence of a dog whistle effect. If Trump's expressions of support for the police were, at heart, a dog whistle intended to appeal to people who felt threatened by challenges to the racial status quo, then support for the police should be associated with votes for Trump specifically among that population. An interaction indicated exactly this: Support for the police was only associated with vote choice among those with high racial resentment. To be clear, many of those low in racial resentment also felt warmly toward the police, but these views were not connected to vote choice.

The connections between vote choice and perceptions of police racial bias and support for BLM also depended in interesting ways on racial views.1515 Although our analyses reflect the role of racial resentment and perceptions of Black political power among the full population, restricting the sample to just non‐Hispanic White voters resulted in substantively identical findings for both the main and interactive effects of these measures of racism. Smaller numbers of respondents of other races and ethnicities made cross‐racial comparisons difficult, but these results hold at minimum for non‐Hispanic White respondents. High racial resentment seemed to insulate people from the anti‐Trump effect that support for BLM otherwise had. Perceptions of police bias seemed to matter less for those who already believed that Blacks have too little political power. In contrast, individuals who simultaneously believed that Blacks held too much political power and that the police were now biased against Whites—those who seemed to feel they were, in Hochschild's (2016) description, “strangers in their own land”—were among the most likely to vote for Trump.

Consistent with the results from other research on the 2016 election, our findings indicate racial resentment and concerns about Black political power seemed to be powerfully associated with the decision to vote for Donald Trump. We also discovered a more complicated role for racial resentment in voter turnout. Clinton's racial justice themes and support for BLM seem to have encouraged some Democratic voters with greater racial resentments to stay home on election day—whereas anti‐racist Democrats voted at a very high rate. In fact, this trend seems similar to what happened in 2008 when the Democratic candidate was Black (e.g., Krupnikov & Piston, 2015; Pasek et al., 2009). In contrast, Trump's candidacy seems to have provoked high turnout among racially resentful Republican voters, as well as to have suppressed turnout among Republicans less racially resentful.

These findings have important implications for research on the police as well as for the political future of a racial social movement concerned with police practices in the United States.

3.4 Implications for understanding views of the police

Criminologists and other social scientists have long been interested in the way the public views the police. This study has several important implications for this body of scholarship. Many scholars have sought either to explain how people view the police or to examine the consequences of such views. For example, researchers have suggested that a lack of faith in the police can limit cooperation with the police, harm informal and formal efforts to control crime, and increase both crime and fear of crime (e.g., Anderson, 1999; Bobo & Thompson, 2006; Drakulich, 2013; Drakulich & Crutchfield, 2013; Kirk & Matsuda, 2011; Kirk & Papachristos, 2011; Sunshine & Tyler, 2003). Our findings indicate that views of the police may also have political consequences. Views of the police were related to both political participation and vote choice in a national election and, of course, may also be relevant to political behavior in other contexts, as well as to the formation of policy preferences, as the findings from a limited number of prior studies have indicated (e.g., Matsueda & Drakulich, 2009). Therefore, political scientists and others interested in political behavior need to pay closer attention to public views of the police.

Indirectly, the idea of the political and symbolic role of the police has other implications for research on views of the police. Research aimed at explaining variation in attitudes toward the police has predominantly been focused on race, socioeconomic status, neighborhood context, criminal victimization, and experiences with police officers (Brown & Benedict, 2002). In particular, prior research has often been focused on the extent, nature, and quality of contact with police: People and communities of color are much more likely to experience coercive contact with the police, and these contacts undermine trust in the police (e.g., Carr, Napolitano, & Keating, 2007; Peffley & Hurwitz, 2010; Rios, 2011; Vargas & Scrivener, 2018; Weitzer & Tuch, 2005a, 2005b, 2006).

From the perspective of procedural justice studies, individuals with more positive views of the police are likely those who have avoided this kind of negative contact with the police. But how should we understand support for the police when people have been exposed to substantial news coverage of police shootings and a social movement accusing the police of acting in racially biased ways? How should we understand this support when politicians use the police as a racial dog whistle? In short, research on views of the police should be focused on considering the social and political symbolic role of the police. It may be necessary to understand positive views of the police as the product not just of positive interactions or a lack of negative experiences but, in some cases, as the product of racial stereotypes about crime or anti‐Black views (see Barkan & Cohn, 1998; Carter & Corra, 2016; Carter et al., 2016; Matsueda & Drakulich, 2009; Mullinix & Norris, 2018). Although procedural justice theory implies that White Americans may be benignly ignorant of racially biased policing, theories of modern racism indicate that White Americans may instead be either willfully ignorant—choosing not to acknowledge injustice that challenges their belief in a just society—or knowingly complicit (e.g., Bonilla‐Silva, 2018). More research on this potentially important distinction is needed.

Additionally, work on legal cynicism and legal estrangement (Bell, 2017; Kirk & Papachristos, 2011) should be designed to consider the potential for even greater marginalization from the police when “support for the police” becomes associated with a particular political party and is used as a racialized dog whistle for those who hope to see the police engage in racial control. This politicization should be especially concerning to the many police chiefs and officers who are trying to improve trust and relations with the communities of color they serve. In short, researchers looking to understand views of the police should better incorporate the racial and political symbolic meaning of the police.

Of course, some criminological work has been focused on the political and racial meaning of the police. In many studies in which the intersection of systemic racism and law and order in American politics was examined, scholars predominantly used archival and historical methods, however (e.g., Alexander, 2010; Beckett, 1997; Hagan, 2010; López, 2014; Murakawa, 2014; Scheingold, 1984, 1992; Schoenfeld, 2018; Tonry, 2011; Weaver, 2007). The macro‐level scope of this research leaves the micro‐level processes by which voters incorporate information about law‐and‐order issues into their political choices unspecified. We recommend that this literature and research on individual‐level views of the police be better integrated.

3.5 Conclusion: A postracial era?

Racial civil rights issues were relevant to the 2016 election: Many voters seemed motivated by the conviction that Black lives matter. In addition to the results reported here, the responses from a series of surveys conducted since the beginning of the BLM movement indicate that awareness of racially disparate policing has increased even among White Americans, that White Americans increasingly agree that society needs to do more to ensure equal rights, and that even Republicans believe President Trump has worsened race relations (Pew Research Center, 2014, 2015, 2017).

At the same time, the results of this research seem to confirm the suggestion that there is a festering reserve of “ressentiment”—racially charged anger about the state of the country directed at, among others, Black citizens. The Civil Rights Movement in the twentieth century gave conservative candidates the opportunity to activate and tap into voters’ concerns about the changes to the racial order these movements presented—often by using crime and justice issues as racial dog whistles (e.g., Beckett & Sasson, 2004; Tonry, 2011). Our results indicate this was again the case in the 2016 election, in which “support for the police” seemed to be a signal that mattered particularly to voters with high levels of racial resentment. This finding reaffirms theories of colorblind and laissez‐faire racism (Bobo & Smith, 1998; Bobo et al., 1997; Bonilla‐Silva, 2018) and shows that a subset of Americans may be cloaking their concerns about the racial order behind a superficially nonracial support of the police.

These findings also put the supposed “postracial” era in a new light. After decades of declining rates of explicit racism (Schuman, Steeh, Bobo, & Krysan, 1997), a lack of racial law‐and‐order rhetoric among general election campaigns between 2000 and 2012, and amidst the immense popularity of presidential candidate Barack Obama in 2007 and 2008, talk began of a new postracial era—a narrative Obama himself embraced (Obama, 2008).

Many scholars, however, warned of the persistence and influence of less explicit indicators of racist sentiment (e.g., Bonilla‐Silva, 2015, 2018; Drakulich, 2015a, 2015b). It is now clear that this postracial era was illusory—a view even Obama now shares (Rupert, 2017). Instead, the era may simply have been one in which racially progressive talk was made possible by the lack of significant threats to the racial status quo. The emergence of a direct threat—in the form of a social movement focused on racial inequalities—may have also caused a more explicit racism to resurface, with consequences for American politics and society more broadly.

APPENDIX A: COEFFICIENTS FROM INTERACTIONS

   

	Turnout				Vote for Trump
Variable	b	SE	b	SE	b	SE	b	SE
Intercept	–4.39***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.72	–2.02**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.60	–.54	1.96	–7.92***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	1.58
Female	.23*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.10	.26**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.10	.20	.14	.24	.14
Age	.03***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00	.03***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00	.00	.00	.01	.00
Married/partner	.16	.11	.19	.11	.26	.18	.26	.19
Separated/divorced/widowed	–.50***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.14	–.49***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.14	.16	.22	.21	.22
# of children in household	–.10*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.04	–.10*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.04	–.08	.07	–.06	.07
Education	.16***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.02	.16***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.02	–.09**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.03	–.09**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.03
Income (in $1Ks)	.01***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00	.01***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00	.00*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00	.00*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00
Unemployed	–.44*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.17	–.48**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.17	.09	.30	–.01	.31
Evangelical/born again	.07	.12	.03	.12	.36*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.16	.36*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.16
Black	.14	.18	.31	.17	–1.25***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.33	–1.08***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.32
Asian	–.39	.29	–.42	.28	–.44	.44	–.38	.46
Other race	–.56**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.20	–.52**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.20	–.78*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.33	–.78*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.34
Hispanic	–.21	.15	–.23	.14	–.57*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.25	–.56*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.25
Foreign born	–.42*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.17	–.39*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.17	–.26	.29	–.28	.30
Face to face	.37**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.11	.36**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.11	.20	.15	.22	.15
Conservative	.02	.04	.07	.04	.42***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.07	.42***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.07
Republican	.11	.14	–.53***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.09	.68***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.04	.69***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.04
Warm toward police	.00	.00	.01***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00	–.04**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.01	.01	.01
Police bias	.20*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.08	.10*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.04	–.26	.24	.10	.18
Warm toward BLM	.02***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00	.00	.00	–.04**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.01	–.01	.01
Racial resentment	–.10	.06	–.66***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.11	–.67	.52	.68***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.09
Black political threat	–.10	.09	–.09	.17	.35**^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.12	1.77*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.75
Warm police × Republican	.00*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00
Police bias × Republican	–.03	.02
Warm BLM × Republican	–.00***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00
Resentment × Republican			.15***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.02
Political threat × Republican			.00	.04
Warm police × resentment					.01***^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00
Police bias × resentment					.00	.07
Warm BLM × resentment					.01*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.00
Warm police × political threat							.00	.01
Police bias × political threat							–.21*^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).	.10
Warm BLM × political threat							–.01	.00

• Notes: N for turnout = 3,233; N for vote for Trump = 2,501.
• ^*p < .05; ^**p < .01; ^***p < .001 (two‐tailed).

Biographies

• Kevin Drakulich is an Associate Professor in the School of Criminology and Criminal Justice at Northeastern University. His research broadly focuses on perceptions of race, crime, and justice both within communities and in broader social and political contexts.

• Kevin H. Wozniak is an associate professor of sociology at the University of Massachusetts Boston. He studies public opinion and the politics of criminal justice.

• John Hagan is the John D. MacArthur Professor of Sociology and Law at Northwestern University and the American Bar Foundation. His research focuses on international criminal law, the effects of parental incarceration on children, and, most recently, on the historical persistence of issues of inequality, crime, and criminal justice in Chicago.

• Devon Johnson is an Associate Professor of Criminology, Law and Society at George Mason University. Her recent research examines public opinion toward punishment and perceptions of police legitimacy in the United States and the Caribbean.