Threat, emboldenment, or both? The effects of political power on violent hate crimes* Received: 20 June 2019 Revised: 14 July 2020 Accepted: 15 July 2020 DOI: 10.1111/1745-9125.12259 ARTICLE Threat, emboldenment, or both? The effects of political power on violent hate crimes* Laura Dugan1 Erica Chenoweth2 1 Department of Criminology and Criminal Justice, University of Maryland—College Park 2 Harvard Kennedy School, Harvard University Correspondence LauraDugan,DepartmentofCriminol- ogyandCriminalJustice,Universityof Maryland,2220LeFrakHall,7251Preinkert Drive,CollegePark,MD20742. Email: ldugan@umd.edu *Additionalsupportinginformationcan befoundinthefulltexttabforthisarticle intheWileyOnlineLibraryathttp:// onlinelibrary.wiley.com/doi/10.1111/crim. 2020.58.issue-4/issuetoc. Wegratefullyacknowledgethehardwork byDarenG.FishertoassurethattheGATE datawereaccurately,validly,andreli- ablycollectedandcodedbytheinterns hesupervised.Hisendlessandcreative endeavorstoaccessopen-sourcemate- rialmadethisresearchpossible.Weare alsogratefulforthe39STARTinternsand MichelleFabianiwhoworkedtirelessly tocodetheGATEdata.Furthermore, wethankPhilSchrodtforhiscontinued technicalsupport.Additionally,weare gratefultoMinXieandtheastuteanony- mousreviewerswhohelpedtomakethis abetterpaper.Finally,wewouldliketo dedicatethisresearchtothelateStephen E.Fienberg,whomentoredtheleadauthor instatistics,andhiswife,thelateJoyce Abstract How do expressions of support or opposition by the U.S. federal government, influence violent hate crimes against specific racial and ethnic minorities? In this article, we test two hypotheses derived from Blalock’s (1967) conceptualization of intergroup power contests. Thepolitical threathypothesispredictsthatpositivegov- ernment attention toward specific groups would lead to more hateful violence directed against them. The emboldenment hypothesis predicts that negative gov- ernment attention toward specific groups would also lead to more hateful violence directed against them. Using combined data on U.S. government actions and federal hate crime statistics from 1992 through 2012, vector autoregression models provide support for both hypotheses, depending on the protected group involved. We conclude that during this period, African Americans were more vulnerable to hate crimes motivated by polit- ical threat, and Latinx persons were more vulnerable to hate crimes motivated by emboldenment. KEYWORDS emboldenment, government actions, political threat, violent hate crime This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. © 2020 The Authors. Criminology published by Wiley Periodicals LLC on behalf of American Society of Criminology 714 wileyonlinelibrary.com/journal/crim Criminology. 2020;58:714–746. https://orcid.org/0000-0001-5930-6950 https://orcid.org/0000-0003-3113-2439 mailto:ldugan@umd.edu http://onlinelibrary.wiley.com/doi/10.1111/crim.2020.58.issue-4/issuetoc http://onlinelibrary.wiley.com/doi/10.1111/crim.2020.58.issue-4/issuetoc http://onlinelibrary.wiley.com/doi/10.1111/crim.2020.58.issue-4/issuetoc http://creativecommons.org/licenses/by-nc-nd/4.0/ https://wileyonlinelibrary.com/journal/crim Dugan and Chenoweth 715 Fienbergwholostherlifetoaperpetra- torofhateintheTreeofLifeSynagogue inPittsburgh,PA,onOctober27,2018. FundingwasprovidedbytheDepartment ofHomelandSecurityScienceandTech- nologyDirectorate’sOfficeofUniversity Programsthroughawardnumber2012- ST-061-CS0001,aswellasbytheCenter fortheStudyofTerrorismandBehav- ior(CSTAB)2.6,whichwasawardedfor STARTtoinvestigatetheunderstanding andcounteringofterrorismwithinthe UnitedStates.Theviewsandconclusions containedinthisdocumentarethoseof theauthorsandshouldnotbeinterpreted asnecessarilyrepresentingofficialpoli- cies,eitherexpressedorimplied,ofthe U.S.DepartmentofHomelandSecurityor START. Fundinginformation ScienceandTechnologyDirectorate, Grant/AwardNumber:2012-ST-061- CS0001 1 INTRODUCTION Since the election of Donald Trump as President of the United States, the media has reported an increaseinthenumberofhatecrimesagainstmarginalizedgroupsdefinedbytheirrace,ethnicity, sexual orientation/identity, and religion (Burch, 2017; Goldman, 2017; Hauslohner, 2018; Potok, 2017).1 In fact, the Center for the Study of Hate and Extremism found that hate crimes reported to the police rose 12.5 percent from 923 in 2016 to 1,038 in 2017 in the ten largest U.S. cities, with the most common motivations being anti-Black, anti-Semitic, anti-gay, and anti-Latino (Levin & Reitzel, 2018). Although hate crime rates have typically spiked during election years in the United States, 2017 was the first year in which they continued to rise once the presidential election was over (Levin & Reitzel, 2018). Indeed, according to the FBI’s annual Hate Crime Statistics report, hate crimes were 17 percent higher in 2017 than in 2016 (7,175 and 6,125, respectively).2 Scholarly analyses have claimed that President Trump’s hostile rhetoric toward specific groups has emboldened individuals to perpetrate violence against those he mentions (Müller & Schwarz, 2018). Although this may be true, we argue that the rise and fall of violence against marginal- ized groups have long been sensitive to gains and losses of political power for those groups (e.g., Cunningham, 2012; McVeigh, Cunningham, & Farrell, 2014; Van Dyke & Soule, 2002), and that the emboldenment of perpetrators toward violence is just one possible dynamic. When we con- siderBlalock’s(1967)power-threattheory,whichexplicitlylinkspoliticalpowertohostilityagainst 1 We use the words “minority,” “marginalized,” and “protected group” interchangeably to refer to nondominant groups. 2 The reports are found exclusively online, last accessed May 25, 2020 (https://www.fbi.gov/services/cjis/ucr/ publications#Hate-Crime%20Statistics). https://www.fbi.gov/services/cjis/ucr/publications#Hate-Crime%20Statistics https://www.fbi.gov/services/cjis/ucr/publications#Hate-Crime%20Statistics 716 Dugan and Chenoweth marginalized groups, we also anticipate that when marginalized groups gain political power, the established dominant group will become more hostile (Van Dyke & Soule, 2002). For instance, it was immediately after Emancipation in 1865 that the Ku Klux Klan formed as a means to pre- serve white supremacy (Levin, 2002). Similarly, when members of the Black community3 gained economic status in various U.S. cities, the communal backlash against them was often severe. In Tulsa, Oklahoma, for instance, White mobs rioting against the Black community left up to 300 dead and tens of thousands homeless (Ellsworth, 1992; Oklahoma Commission to Study the Race Riot of 1921, 2001). Countless other examples in U.S. history demonstrate this pattern. Prior research typically analyzes the association between political threat and various forms of nonviolent reactionary mobilization, such as social movement formation (Cunningham, 2012; Van Dyke & Soule, 2002) or dominant group voting behavior (McVeigh et al., 2014). But our study turns to Blalock’s (1967) original conceptualization of intergroup power contests—and what became known as “power-threat theory”—to articulate political threat and emboldenment hypotheses that explain changes in patterns of violent hate crimes against specific groups in the United States since the early 1990s. Although all hate crimes are inherently violent, we explicitly focus on violent hate crimes using the Federal Bureau of Investigation (FBI) definition because of the extremity of the acts, because of the enduring relevance of direct violence as a tool by which dominant groups intimidate and control marginalized groups (Müller & Schwarz, 2018; Tolnay & Beck, 1992), and because measures of violent crimes are less subject to underreporting bias than nonviolent crimes (Hart & Rennison, 2003). The political threat hypothesis predicts that violent backlash against specific groups is triggered by political gains made by those groups (e.g., new civil rights protections, meaningful changes in economic and social status, or support among powerful political elites; Taylor, 2016). This inter- pretationwouldcharacterizeviolenthatecrimeasamechanismtoreinforcethedominantgroup’s dominion in the power equilibrium across groups (Blalock, 1967). In this way, perpetrators asso- ciated with dominant groups use violent hate crimes to try to reestablish social dominance when marginalized groups begin to achieve more power (Eitle, D’Alessio, & Stolzenberg, 2002). If polit- ical threat contributes to the variation in violent hate crimes, we would expect to see increases in violent hate crimes against groups that recently gained political power, signaling a potential threat to status quo hierarchies. A complementary approach, which we call the emboldenment hypothesis, predicts that increases in violent hate crimes against certain populations are triggered by government elites who signal supremacy over those groups, emboldening some members of the dominant group to commit violent action. Within the context of intergroup power, we interpret explicit anti-group acts by the government as signals to some members of the dominant group that they can combat the perceived threat of minority power growth without consequence. In other words, some may perceive that the government is giving them a license to act on their anger. Indeed, scholars have tied hate speech—particularly by those in power—to discriminatory action, retaliatory violence, and hate crimes (Gagnon, 1995; Kalmoe, 2014; Kteily, Bruneau, Waytz, & Cotterill, 2015; Müller & Schwarz, 2018). As Sara Lipton argued, “[A] heightening of rhetoric against a certain group can incite violence against that group, even when no violence is called for. When a group is labeled hostile and brutal, its members are more likely to be treated with hostility and brutality” (2015, p.1).Forinstance, formanydecadesaftertheendofdejureslaveryintheUnitedStates,politicians andlawenforcementofficialscondonedthelynchingofBlackAmericans,signalingimpunityand approval for those who would engage in such acts (Bouie, 2017; Tolnay & Beck, 1995). 3 We use the terms Black and African American interchangeably in this article. Dugan and Chenoweth 717 Most existing studies of political threat have been focused on the expansion of wealth, voting, population growth, or political representation among marginalized groups as the main source of political threat (Blalock, 1967; Eitle et al., 2002; Taylor, 2016; Van Dyke & Soule, 2002). But we argue that focused attention from politicians—through their speech and their policy proposals— can induce backlash or emboldenment effects without necessarily creating material benefits or committingmaterialharmtomarginalizedcommunities.Evenwithoutalteringdefactoeconomic opportunities or political power, favorable actions by the federal government may signal the gov- ernment’s commitment to expanding economic and agenda-setting power for African Americans or other marginalized groups. This expectation may result in anticipatory mobilization, mean- ing that backlash could occur without any favorable policies coming to pass. Conversely, focused negative government attention—through officials’ discriminatory or racist speech and policy proposals—might similarly embolden violent hate crimes against the relevant group. Indeed, prior research on emboldenment has often been focused on the ways in which official rhetoric— such as racist or exclusionary language—can propel violence against marginalized communities (Gagnon, 1995; Kalmoe, 2014; Kteily et al., 2015; Lipton, 2017; Müller & Schwarz, 2018). Thus, both policies and rhetoric regarding a protected group can produce political threat, emboldenment, or both effects. Our study therefore assesses these two compatible components of power threat theory, which predict an increase in violent hate crimes both during times of increased political power for spe- cific marginalized groups (political threat) and during times of political retrogression for such groups (emboldenment). We generate hypotheses and test them using vector autoregression on monthly data constructed from recorded events on U.S. federal government actions—including speech and material action—and violent hate crimes specific to Black and Latinx people between 1992 and 2012. Our aim is to assess how political threat and emboldenment relate to changes in violent hate crimes targeting those groups. Although we find different effects across each demo- graphic group, our findings suggest that speech and actions by federal officials toward these groups affect patterns of hate crimes against them. We conclude that during this period, African Americans were more vulnerable to hate crimes motivated by political threat, and Latinx people were more vulnerable to hate crimes driven by emboldenment. 2 INTERGROUP POWER CONTESTS—POWER THREAT THEORY AtatimeofheightenedracialdiscordintheUnitedStates,Blalock(1967) introducedwhatbecame knownas“power-threat theory”in hisbookTowardaTheoryofMinorityGroupRelations. Blalock characterized intergroup power contests by the shifts away from and toward a power equilib- rium where the dominant group maintains greater power relative to the minority group (see also Blumer’s 1958 explanation of group position). Although the theory is meant to be generalized to any set of groups where one dominates over the others, the book and subsequent writings on this topic were focused mostly on Anglo- and African-American majority and minority groups, respectively (e.g., Fossett & Kiecolt, 1989; Giles & Evans, 1986; Jacobs & Helms, 1999; Quillian, 1995, 1996; Taylor, 1998). Blalock (1967) argued that the dominant group in a particular setting acts to maintain a power equilibrium relative to a minority group, and that these power shifts are func- tions of increases in resources and mobilization efforts: “...resources seem to depend primarily on the motivation and goals of persons over whom power is exercised, whereas mobilization is more largely a function of the goals of the persons exerting the power” (Blalock, 1967, p. 114). In other words, as a minority group gains resources, the dominant group mobilizes its resources to 718 Dugan and Chenoweth reestablish its dominance within the power hierarchy (i.e., the political threat hypothesis). This dynamic has been documented by Wacquant (2006) and others, who have shown that Jim Crow laws supplanted slavery, and Western and Pettit (2010), who argued that mass incarceration of African Americans supplanted Jim Crow laws (see also Alexander, 2012). Van Dyke and Soule (2002) similarly showed how increases in perceived political, economic, and demographic threats toWhitesupremacyledtoincreasednumbersofWhitenationalistandWhitesupremacistmilitias inthe1990s.Moreover,Vazquez(2010)arguedthatbyincorporatingimmigrationlawintotheU.S. criminal justice system, dominance was reestablished over Latinx people following major gains in labor and immigration rights. Although the key variables used by scholars in this area are minority resources and dominant group mobilization, Blalock’s (1967) conceptualization of relative power includes resources and mobilization for both groups. Of course, the underlying assumption here is that the equilibrium, as defined by the dominant group, is final. In actuality, systems that preserve one group’s domi- nanceoverothersareincreasinglyinterpretedbysomemembersofthepublicasinherentlyunjust and discriminatory, making them vulnerable to legal challenges and other forms of change over time (Epps, 1998). Theseshifts inperspectivetendtoresult frommobilizationbymarginalizedgroups.Suchmobi- lization is epitomized by the abolition movement with regard to slavery; the civil rightsmovement in response to Jim Crow and racial segregation; and the women’s and lesbian, gay, bisexual, trans- gendered, and queer (LGBTQ) movements in response to laws restricting the rights of women and sexual minorities (Van Dyke & Soule, 2002). Scholars have referred to periods of mobilization that result inmajorgainsforaggrievedgroupsasthe“rightsrevolutions”(Epps,1998).Wecharacterize minority group mobilization as a persistent effort to destabilize the dominant group’s equilibrium of inequality. Just as minority group mobilization is persistent, dominant group resources are plentiful as default narratives have historically favored members of the dominant group, which since at least the nineteenth century in the United States have been White, male, natively born, heterosexual Christians (Doane, 1997). But the dominant group is not monolithic. Instead, it is vulnerable to fragmentation as mobilization and resistance by marginalized groups begin to challenge the sus- tainability of the existing order (Epps, 1998). For example, Whites in the United States split over the abolition of slavery, resulting in a civil war between Union and Confederacy forces. A similar dynamic occurred in South Africa, where White business elites split over the continuance of the apartheid system (Charney, 1999; Fourie & Eloff, 2005). In the twenty-first century United States, immigration is a notably partisan issue, with Whites divided mainly along political party lines concerning appropriate approaches to immigration and immigrants (Pew Research Center, 2018). We argue that the constant threat of fragmentation tends to tame overt expressions of hostility toward marginalized groups, as the dominant group publicly acclimatizes to a growing sympathy among the broader population. Despite having ample resources to reestablish dominance, threat- ened Whites feel regular pressure from minority group resistance and other Whites to accept new equilibria, tempering the capacity to mobilize all available resources to reestablish social domi- nance. It is under this tension that any overt hostile act by an elite member of the dominant group could motivate violence against marginalized groups (i.e., the emboldenment hypothesis). In this article, we focus on Black and Latinx individuals in the United States. We argue that key metrics of federal power for any minority group include material changes implemented at the federal level, as well as public support expressed by federal officials—like the Voting Rights and Civil Rights Acts, which rendered Jim Crow laws illegal; or the DREAM Act, which expressed support for deportation immunity to immigrants who arrived in the United States as Dugan and Chenoweth 719 children. We argue that such changes only result from minority group mobilization and create new resources for ongoing mobilization efforts. For instance, pressure from the Civil Rights movement was mainly responsible for the eradication of legalized segregation and Jim Crow, even as key leadership of the movement negotiated with federal officials regarding the terms of the Civil Rights and Voting Rights Acts. And the Dreamer movement, which mobilized hundreds of thousands of undocumented immigrants to defer deportation and allow undocumented students to attend college in the United States, was responsible for the passage of Deferred Action for Childhood Arrivals (DACA) and for increasing support for immunity for illegal immigrants (Pew Research Center, 2018). Yet, such accommodations can provoke backlash as well, as others fear disempowerment by the perceived power growth of minority groups. Bouie (2017) argued that expressions of political racism have been cyclical as African Americans experienced different levels of political gain and loss across different time periods. For instance, the landmark case Brown v. Board of Education (1954) ruled that racial segregation of children in schools was unconstitutional and mandated that they be integrated. Initially hailed as a major step toward justice, the ruling also fueled backlash from pro-segregationists, with some governors defying federal orders to integrate (Fobanjong, 2003). More recently, after the passage of DACA, the Obama administration ramped up deportation of Latin American immigrants, and the Trump campaign tapped into anti-immigrant sentiment by referring to Mexicans as “drug dealers,” “criminals,” and “rapists” and promised to build a border wall to end illegal immigration (Phillips, 2017), rhetoric that he has maintained as president. Therefore, publicly expressed disdain toward minority groups by members of the federal government provides the dominant group a new resource that it can use in its efforts to return to a status quo equilibrium, including an expectation of impunity for crimes committed against the group. Combined,wehypothesizethat increasesinexpressedfederalsupport forspecificmarginalized groups puts members of those groups at risk for violent hate crime victimization (political threat), andthat increasesinfederaloppositiontoprogressforspecificmarginalizedgroupsalsoputstheir members at risk of violent hate crimes (emboldenment). The following sections further delineate the causal mechanisms linking U.S. federal actions to violent hate crime for each hypothesis. 2.1 Political threat hypothesis Blalock (1967) explained that as minority resources increase and are mobilized, they increase political power for marginalized groups and thereby threaten the dominant group’s position of power. This leads the dominant group to mobilize, which in turn reestablishes the power balance preferred by the dominant group. In the same vein, we would expect that evidence of increases in minority mobilization would also raise concerns of possible lost relative power for the dominant group. Empirical evidence has primarily supported this relationship through various operationaliza- tions of minority resources and dominant group mobilization. A commonly used metric of minor- ity resources is the percentage of the population that is Black, which has been found to be associ- ated higher levels of anti-Black prejudice (Quillian, 1995; Taylor, 1998), support for harmful racial policies (Fossett & Kiecolt, 1989; Giles & Evans, 1986; Quillian, 1996), increase in Republican voter registrations in Louisiana (Giles & Hertz, 1994), votes for Wallace in 1968 (Wright, 1977), spending on corrections (Jacobs & Helms, 1999), spending on police (Jackson & Carroll, 1981), support for the death penalty (Baumer, Messner, & Rosenfeld, 2003), and support for other tough-on-crime 720 Dugan and Chenoweth policies (King & Wheelock, 2007). Of particular relevance to the current research is King’s (2007) conclusion that law enforcement in jurisdictions with large Black populations is less likely to comply with federal hate crime law (i.e., classify crimes as hatefully biased). We will return to this observation later as it has implications for the dependent variable in this study. Other scholars measure racial threat through mobilization, such as rioting or protest, and find it associated with spending on corrections (Jacobs & Helms, 1999), support for establishing “law andorder”(Wasow,2020),andpoliceexpenditures(Jackson&Carroll, 1981).Behrens,Uggen,and Manza (2003) measured minority threat by the percentage of the size of the prison population, based on the logic that dominant group mobilization is measured by felon disenfranchisement laws. In other words, as people leave prisons, they will be unable to vote in elections, which dis- proportionately diminishes Black political power because of the disproportionate representation of Black people in prison. Racial threat hypotheses have also been tested using a percentage of the population that is immigrant, finding an effect on the arrest rate for drunkenness (Brown & Warner, 1992) and prejudice against immigrants (Quillian, 1995). Jacobs and Carmichael (2001) found that Black and Hispanic population growth is associated with the growth of the prison population for each group. Finally, Green, Strolovitch, and Wong (1998) found that when minori- ties (Asians, Latinos, and African Americans) moved into New York neighborhoods with White strongholds, hate crimes against these groups increased. This last study most closely aligns with the current research as we also measure dominant group mobilization as a rise in hate crimes targeting specific groups. 2.2 Emboldenment To the best of our knowledge, no other scholarship has framed emboldenment within the con- text of power-threat theory or intergroup power contests. The long-running struggles between the dominant group and different racial, ethnic, religious, and sexual minorities in the United States, however, suggest that mobilization by marginalized groups deserves attention in any study of intergroup power contests. Although dominant group resources remain plentiful, they are also restrained by the risk that their overuse will draw backlash from other dominant group mem- bers, inadvertently empowering targeted minority groups (Dugan, Huang, LaFree, & McCauley, 2008).Inotherwords, ifdominantgroupmembersopenlyattackminorities,othersmightdistance themselves from the dominant group and could even support the mobilization efforts by minori- ties. Therefore, dominant resources are more than just the ability to reestablish dominance, but they also entail the ability to do so without alienating other dominant group members. Thus, continuous pressure of minority mobilization combined with the tempered dominant group resources creates a potential increase of dominant group mobilization when a revered member of the dominant group legitimizes minority group repression. Thus, the emboldenment hypothesis suggests that public action by an established authority against a specific minority group raises the level of dominant group resources enough for some members of the dominant group to counter-mobilize against persistent minority group mobilization. Indeed, Müller and Schwarz (2018) found a rise in anti-Muslim hate crimes after Presi- dent Trump’s tweets on Islamic-related topics—tweets that have typically been hostile toward Muslims—suggesting that the violence was emboldened by Trump’s rhetoric. Kalmoe (2014) tested this relationship more directly by exposing subjects to a political ad that conjured either violent metaphors or nonviolent synonyms. Those with aggressive traits were more likely to sup- port political violence when exposed to the violent metaphors. This provides ostensible support Dugan and Chenoweth 721 for the emboldenment hypothesis for those members of the dominant group who might already feel heightening threats from marginalized groups. If true, the emboldenment hypothesis sug- gests that political elites have the power to exacerbate tensions across groups. In fact, Gagnon (1995) argued that elites are aware of their power and that they intentionally provoke violent con- flict between the dominant group and others because such conflict tends to focus attention exclu- sively on ethnic identity and away from other politically charged topics that could compromise their power. This argument is also made by Tolnay and Beck (1995), who claimed that the White elites of the Antebellum South benefited from the antagonism between African Americans and poor Whites as it kept them from forming a coalition of laborers. By fostering this antagonism, White elites allowed White mob violence that then led to thousands of lynched African Ameri- cans. Indeed, the authors hypothesized that the eventual decline in U.S. lynchings was a result of the discouragement of the White elite, who were trying to entice low-wage Black laborers to remain in the South during the Great Migration (Tolnay & Beck, 1995). In sum, we measure minority and dominant group resources as the number of pro- and anti- group actions by the U.S. federal government each month from 1992 through 2012, respectively. We focus this research on African Americans and Latinx persons because there was significant public attention to movements against racial discrimination and immigration policy, especially regarding immigration from Latin America, during this period. 3 HYPOTHESES The political threat hypothesis expects that as political elites take actions to further empower marginalized groups, violent hate crimes against them would increase. We generalize this approach by considering whether the growth in power of African American and Latinx popu- lations elicits the same response from the dominant group. Combined, the following hypothesis comprise our expectations according to political threat theory. Hypothesis 1. Federal government actions that favor Black or Latinx persons are related to increases in violent hate crimes targeting Black and Latinx persons, respectively. The emboldenment hypothesis expects that as federal officials take actions that undermine marginalized groups, a subsequent increase in violent hate crimes against those groups follows. Once again, we generalize this approach by considering whether federal government actions opposing African American and Latinx persons elicits increases in hate crimes from the domi- nant group. Combined emboldenment leads to the following hypothesis: Hypothesis 2. Federal government actions that oppose Black or Latinx persons are related to increases in violent hate crimes targeting Black and Latinx persons, respectively. 4 DATA AND METHOD 4.1 Data The primary data come from two different sources that chronicle individual events and can there- fore be aggregated to any temporal unit. The dependent variables measure violent hate crimes in 722 Dugan and Chenoweth the United States, which were compiled by the FBI’s Uniform Crime Reporting (UCR) program for its Hate Crime Database under the Hate Crime Statistics Act of 1990 (HCSA). The primary independent variables come from the Government Actions in Terror Environments-United States (GATE-USA) data. All events are aggregated to the month and combined for analysis beginning in January 1992, the first month that the UCR hate crime data are available, and ending in Decem- ber 2012, the last month that GATE-USA data are available, totaling 252 months. Other data come from public sources that provide the congressional sessions when each party held the majority in both houses, the number of tax-exempt civil rights or advocacy organizations for each year, and other UCR data. Also, supplemental analyses in appendix E in the online supporting information use data from the National Crime Victimization Survey (NCVS).4 4.1.1 Violent hate crime rate The HCSA of 1990 authorized the FBI’s UCR program to collect data on “crimes that manifest evidence of prejudice based on race, religion, sexual orientation, or ethnicity” (U.S. Department of Justice, 2009, p. 4). Data collection follows a similar protocol as other UCR data sets, which relies on voluntary reporting of each law enforcement agency to a centralized state repository. However, instead of submitting reports every month, agencies either submit reports every quarter that list details on each incident through an incident report or submit a zero report verifying that no hate crimes were reported to the agency over the last quarter. NIBRS-compliant agencies can forgo these reports by simply marking an indicator that the documented event was a hate biased crime. Despite the similarities in reporting for crimes with and without hateful bias, without fed- eral money allocated to help local agencies with their additional administrative and investigatory burdens,participatinghasbeentenuous—especiallyinthebeginning(McVeigh,Welch,&Bjarna- son, 2003). In 1991, the population represented by participating agencies was less than 50 percent; however, that number rose to 85 percent by 2002 (King, 2007). The absence of additional investigatory support is especially detrimental to compliance. To report a bias in the HCSA, law enforcement must investigate the reasons behind a crime rather than just uncovering the material actions that unfolded (Boyd, Berk, & Hamner, 1996). Without resources for specialized investigations, many departments simply report no hate crimes (Jenness &Grattet,2001;McVeighetal.,2003).Evenwhenmotivesareapparentwithoutadditionalinquiry, different officers interpret similar circumstances differently (Martin, 1995). Furthermore, the con- troversial nature of assigning bias to a perpetrator sometimes makes both victims and officers reluctant to report a crime as hate motivated (Cronin, McDevitt, Farrell, & Nolan, 2007; McVeigh et al., 2003). In fact, because hate crimes are often the result of intergroup conflict, labeling or even failing to label a crime as hate motivated can be interpreted as law enforcement choosing a side, possibly making them vulnerable to protest (Bell, 2002; McVeigh et al., 2003). Concern about biases in the HCSA collection has led scholars to scrutinize the variation in hate crime reporting across jurisdictions. Hate crime data were found less likely to be collected in jurisdictions with larger Black populations (King, 2007) and in the South or Midwest (McVeigh et al., 2003). In fact, McVeigh et al. (2003) interpreted the counties with high numbers of hate crimes as those that have higher compliance (i.e., more support for minorities) rather than simply having more hate crimes. Following this logic, they argued that counties with successful social 4 Additional supporting information can be found in the full text tab for this article in the Wiley Online Library at http: //onlinelibrary.wiley.com/doi/10.1111/crim.2020.58.issue-4/issuetoc. http://onlinelibrary.wiley.com/doi/10.1111/crim.2020.58.issue-4/issuetoc http://onlinelibrary.wiley.com/doi/10.1111/crim.2020.58.issue-4/issuetoc Dugan and Chenoweth 723 movements would have more advocates lobbying for bias crimes to be recorded in the HCSA. Furthermore, King (2007) used minority threat theory—the same theory used in this article—to argue that the more “threatening” jurisdictions are less likely to participate in HCSA. Indeed, he found that jurisdictions in places with large Black populations are less likely to comply with the hate crime law. Given these and other conclusions from years of scrutiny, variation in counts of U.S. hate crimes is likely driven by a mixture of actual events and the ability and willingness of victims and local agencies to report events as hate crimes. Combined,theseconcernsraiseconsiderabledoubtaboutthevalidityofanyresearchthatrelies on the HCSA data as a dependent variable, especially for time series beginning in 1992, like that which is used in the current research. We recognize this challenge and have included several features of this research that are designed to mitigate this problem. Additionally, we conduct sup- plementary analysis using victim reports of hate crimes and subsets of the HCSA agencies that participated for all or most of the years from 1992 through 2012. First, the main analyses use data aggregated across all jurisdictions to the monthly unit, thus, absorbing all geographic variation and relying entirely on temporal variation in the violent hate crime rates for our estimations. By aggregating across jurisdictions, variation in compliance is less ofaconcern.Thisstrategywouldonlymitigateallconcerns,however, ifcompliancewereconstant over the two decades of analysis; and we know that compliance has increased over this period (Cronin et al., 2007; King, 2007; McVeigh et al., 2003). As crime rates are constructed by calculat- ing the frequency of events in the numerator and the total population at risk in the denominator, erroneously assuming constant compliance over the series leads to systematic bias. To demon- strate, if the true rate was flat over the series, but compliance increased, then by using the total U.S. population in the denominator, the rate would appear to increase over the series. To reduce this type of bias, we generate violent hate crime rates using the populations represented only by those agencies that participated in the HCSA during the quarter associated with each month. We identify those agencies that participate in the HCSA following the practice of King (2007) by using data from the batch header files to identify agencies that turned in either an incident or a zero report that quarter (UCR, 2017). These files also include the population represented by the reporting agencies, regardless of whether they participated.5 We anticipate that this overall strategy improves the accuracy of the rates, yet we also recognize that they are still likely biased downward, as some agencies erroneously report no or low incidents of hate crimes, and there- fore, the dependent variables are still measured with error (Cronin et al., 2007). Furthermore, by limiting the population represented in the denominator to only a subset of the total popula- tion,generalizabilitytotheentirenationiscompromised.Weknowthatgeneralizability,however, is already compromised by using data that rely on selective compliance. By using the reporting agency population in the denominator, the variation over time in the newly constructed depen- dent variables more closely mimics the variation in the underlying actual violent hate crime rates than when we use the total population in the denominator. The results presented here use the agency population in the denominator. A second set of results, however, that uses the total pop- ulation in the denominator is reported in appendix D in the online supporting information. The findings are nearly identical. 5 Someparticipatingagenciesarerecordedashavingzeropopulationevenwhentheirpopulationgroupreportsinhabitants (e.g., under 10,000). Therefore, when the agency population is listed as zero, but the population group indicates a positive population, the maximum population value was used as the population for that agency. Although this likely inflates the population to some degree, we view it as the more conservative strategy as it deflates the violent hate crime rate. 724 Dugan and Chenoweth Furthermore, because of the unavoidable selective compliance, we dedicate appendix B in the online supporting information to assessing the spatial variation of compliant agencies across states. This analysis shows that the distribution of agency participation by state is uneven. For example, the largest proportion of participating agencies is in Texas, whereas agencies in Hawaii, Alaska, Georgia, and Alabama show the least participation. Finally, three variables are added to the models to address potential bias in the estimates. We first include the total number of agencies (measured in the thousands) that submitted either an incident or a zero report that quarter. As the denominators of the dependent variables are cal- culated using only the population from these reporting agencies, that denominator is a function of the participating agencies. Furthermore, an agency’s decision to participate in HCSA could be influenced by attention given to protected groups by the federal government, which would make the number of participating agencies an important control to mitigate possible bias. The second control variable is included because only those participating agencies that submit inci- dent reports (rather than zero reports) can influence the numerator in the violent hate crime rates. Furthermore, it is unclear whether zero reports reflect true zeros or the agency’s reluctance to report biased crimes, which could also be influenced by the tone set by the federal government. Thus, we include the percentage of participating agencies that submitted incident reports as a second control. Third, we follow the lead of McVeigh et al. (2003) by adding the number of civil rights organizations (measured in thousands) from the National Center for Charitable Statistics Core Files data sets for all years, which was calculated for each year using the same protocol as McVeigh and colleagues (NCCS, n.d.). This variable approximates efforts by civil rights organi- zations to lobby law enforcement to comply with the HCSA and report biased crimes as such so they can be prosecuted accordingly. Furthermore, because the number of civil rights organiza- tions also approximates minority mobilization, this variable is also theoretically relevant to the hypotheses. Thus, we expect that some of its variation relates to hate crime reporting (McVeigh et al., 2003) and that some of its variation relates to hate crime perpetration as the number of civil rights organizations approximate minority group mobilization. In each case, we would anticipate that the number of civil rights organizations would be positively related to violent hate crimes. Although we expect that these efforts improve the validity of the findings, we recognize that the data are inherently problematic as they systematically ignore hate crimes unreported and/or unrecognized by law enforcement. Furthermore, it is possible that victims’ and agencies’ motiva- tions to report crime as biased are influenced by changes in the federal attention paid to specific groups, above and beyond the controls described earlier. If local agencies are reluctant to report crimesashatefullybiasedwhenthefederalgovernmentisgivingthetargetedgroupeitherpositive or negative attention, then our results will be biased downward. Unfortunately,thesedataarethebestavailableforourpurposes.TheNationalCrimeVictimiza- tion Survey (NCVS), however, did begin collecting data on hate crimes as reported by the victims in 2003. Unsurprisingly, the estimated number of hate crimes generated from victim reports is substantially higher than that reported by police agencies. In fact, between 2005 and 2015, the NCVS estimates around 250,000 hate crimes each year, whereas the HCSA only reported a little more than 7,000 biased crimes in 2015 (Office of Justice Programs, 2018). Although the estimates from the NCVS data are undoubtedly closer to the actual numbers, limitations in the NCVS data set preclude it from being used in the current analysis. For example, because it relies on sampling, not all months have an episode of violent hate crimes against Black or Latinx persons. This gener- ates many missing values at the monthly level. Also, the rates will exhibit high variation, making estimation at that level difficult. Because the NCVS and the HCSA are both measuring the same underlying dynamic, however, a comparison between the two might help validate the HCSA data. Dugan and Chenoweth 725 Supplemental analyses in appendix E in the online supporting information describe in detail the similarities and differences between the two data sources, estimates their trend and first differ- enced correlation, and re-estimates the main models with the HCSA data, the NCVS hate events known to the police, and the NCVS hate events unknown to the police. Finally, as another check on the validity of these data, we repeat the main analyses using rates generated from only the 741 agencies that participated for all four quarters in all 21 years in the series and the 1,973 agencies that participated in all four quarters for 20 of the 21 years. As a point of contrast, the main analyses rely on data generated from 22,213 different agencies that partic- ipated in at least one quarter during the 21 years. Violent hate crime trends against Black and Latinx persons from all three samples are compared prior to analyses. Appendix B in the online supporting information describes the spatial distribution of those agencies in both subsets, as well as in the main dataset. To test the hypotheses, we rely on the details provided in the HCSA data. Each event includes information on the race of the offender (if known), type of victims, bias motivation, offense type, and location type. Because we are only interested in violent hate crimes as defined by the FBI, we retain only murder, non-negligible manslaughter, negligible manslaughter, kidnapping or abduction, forcible rape, forcible sodomy, sexual assault with an object, forcible fondling, rob- bery, aggravated assault, simple assault, and intimidation. Only those hate crimes that were per- petrated by White or unknown offenders were retained as they are members of the dominant group. Finally, only crimes with biases that were anti-Black, anti-multiracial, and anti-Hispanic were retained for the specific analyses. After applying these criteria, we retain 39,599 violent hate crimes against Black people (including anti-multiracial), aggregate them to the month, and calcu- late the rates per 1,000,000 persons living in participating jurisdictions. Data used for the Latinx models include only violent anti-Hispanic hate crimes resulting in 8,004 violent crimes to cal- culate monthly rates per 1,000,000 persons. We use natural logarithms of these rates to improve normality. 4.1.2 Federal government actions relevant to specific groups These data come from a larger GATE project that documents all actions by the federal U.S. gov- ernment that are relevant to political extremists and their and their constituencies’ grievances (e.g., the far-right; see Dugan & Chenoweth, 2017). The original source of the GATE-USA data collection is all Reuters news articles that mention key federal U.S. government actors between 1987 and 2012 (see Dugan & Chenoweth, 2017, for the complete list of actors). Lead sentences were extracted and coded using Textual Analysis by Augmented Replacement Instructions (TABARI), which searches them and identifies stories that match the criteria of an extensive set of dictionar- ies designed to capture political activity (Schrodt, 2012) and supplemented with names of known extremists in the United States (Freilich, Chermak, Belli, Gruenewald, & Parkin, 2014; Smith & Damphousse, 2001; START, 2016). Final cases were coded by research assistants and cleaned by the investigators. As such, each action was reviewed by at least two persons, one of which was a principal investigator to assure consistency across all cases.6 6 Agreement statistics were not calculated because disagreement in earlier parts of the coding protocol led to greater dis- agreement elsewhere. To assure validity and reliability, the coders and investigators had weekly meetings throughout the project to discuss ambiguous cases. All coding decisions were recorded and applied across all cases. 726 Dugan and Chenoweth For the current project, we kept only actions—which include both speech and material acts— byU.S. federalactorsthatwererelevanttoBlackand/orLatinxpersons(includingall immigration -relevant actions) that occurred between January 1992 and December 2012. These cases were col- lected in the original GATE data because of their relevance to grievances by far-right extremists. The final data includes 148 non-neutral actions that were relevant to Black people; 116 that politi- cally benefited them (e.g., prosecuting White supremacists and promoting racial equality) and 32 thatdisadvantagedthem(e.g.,challengingAffirmativeAction).Wealsoselectedonlythoseevents that were relevant to Latinx persons specifically or immigration more generally, resulting in 220 non-neutral actions with 109 supporting Latinx or immigration rights and 111 against such rights. We retained immigration-relevant actions that were specific to non-Latinx persons because those events are set within the larger context of immigration, as the United States public tends to con- flate immigration policy with Latinx-specific immigration (Lopez, Morin, & Taylor, 2010).7 We aggregated all events to the month and according to whether they favored or disadvantaged the protected groups and used their square roots for analysis to mitigate non-normality. When aggregated to the month, the GATE data are intended to capture the amount of positive and negative attention given to specific marginalized groups. This differs from efforts to delineate relevant legislation because it relates more to the public conversation than to the legislative pro- cess. Nevertheless, legislation is captured indirectly because more attention is given to the group during periods when relevant legislation is introduced and voted on (e.g., during efforts to remove Affirmative Action with the introduction of the 1995 Equal Opportunity Act into the U.S. Senate). Also, GATE captures some of the legislative process, as coders were directed to capture all events that reported 1) the promotion of a policy idea prior to proposal, 2) a Bill proposal, 3) a congres- sional subcommittee vote, 4) a vote by the full House or Senate, 5) the signing of a Bill, 6) the veto of a Bill, 7) the override of a veto, 8) an executive order, or 9) responses to a new policy, law, or act. Appendix A in the online supporting information describes the key patterns of federal attention given to each group in the GATE data. 4.1.3 Sources for other key variables As mentioned earlier, the number of civil rights organizations came from The National Center for Charitable Statistics (NCCS, n.d.), which derives its files from databases maintained by the Inter- nal Revenue Service (IRS) on tax-exempt nonprofit organizations. The number of organizations from the NCCS core files with National Taxonomy of Exempt Entities (NTEE) core code classi- fications that indicate civil rights or advocacy organizations (codes R20 through R30). Because these data are only measured once each year, all tallies for months in the same year are the same, consequently deflating estimated standard errors and risking false significance. In fact, this vari- able only has 21 unique values out of the total number of 252 months. For this reason, we consider the estimated effects civil rights organizations to be informative rather than definitive. We also include controls for whether the White House, House of Representatives, and Sen- ate are all controlled by the Democrats or Republicans as the political climate likely affects both government actions and violent hate crimes. Data on the composition of Senate and House of 7 Nevertheless, earlier models that used an immigration variable that excluded actions that were specific to non-Latinx persons produced substantively similar findings, although the broader measure was more strongly related to violence against Latinx persons specifically. Dugan and Chenoweth 727 TABLE 1 All variables included in model with their operationalizations Variables Operationalization Dependent Variables Ln Violent Hate ln ( freq of violent bias crimes n persons living under reporting agency × 1,000,000 ) Government Actions Sq. rt. Pro-Group GA The square root of the number of U.S. federal government actions that give positive attention to specific group in current month Sq. rt. Anti-Group GA The square root of the number of U.S. federal government actions that give negative attention to specific group in current month Control Variables Civil Rights Groups The number of civil rights or advocacy groups, measured in thousands, registered with the IRS as tax exempt Democratic Controlled A dummy variable (1/0) indicating months when the presidency, Senate and House of Representatives are all controlled by the Democratic party Republican Controlled A dummy variable (1/0) indicating months when the presidency, Senate and House of Representatives are all controlled by the Republican party Violent Crime Rate freq of violent crimes n persons living in US × 1,000,000 Total Participating Agencies (PA) The total number of UCR participating agencies, measured in thousands, that submitted either an HCSA incident or zero report to the FBI in the quarter associated with the current month Percent PA with Incid Reports The percent of total participating agencies that submitted an incident report (rather than a zero report) in the quarter associated with the current month January—November Dummy indicators (1/0) for each month from January through November, using December as the reference month Representativememberswerecompiledfromseveralsourcesaccessedinthefallof2017.8 Further- more, because violent crime rates are likely to be driven by violent crime in general, we control for the monthly violent crime rate using the FBI’s UCR data. Finally, to absorb variation caused by seasonal affects, we include dummy variables for each month using December as reference. See table 1 for a list of all variables and their operationalization. 4.2 Methods Because of the concern for how representative the HCSA data are, we begin the analysis by exam- ining the distribution of participating agencies by state. We do this for the entire data set over the 21 years, and then we examine subsets of agencies that more fully participated over the entire period. 8 These include https://www.senate.gov/history/partydiv.htm; https://www.senate.gov/artandhistory/history/common/ briefing/senators_changed_parties.htm; https://www.brookings.edu/multi-chapter-report/vital-statistics-on-congress/; http://history.house.gov/Institution/Party-Divisions/Party-Divisions/; http://history.house.gov/Congressional-Overview /Profiles/1st/; and https://www.brookings.edu/multi-chapter-report/vital-statistics-on-congress/. https://www.senate.gov/history/partydiv.htm https://www.senate.gov/artandhistory/history/common/briefing/senators_changed_parties.htm https://www.senate.gov/artandhistory/history/common/briefing/senators_changed_parties.htm https://www.brookings.edu/multi-chapter-report/vital-statistics-on-congress/ http://history.house.gov/Institution/Party-Divisions/Party-Divisions/ http://history.house.gov/Congressional-Overview/Profiles/1st/ http://history.house.gov/Congressional-Overview/Profiles/1st/ https://www.brookings.edu/multi-chapter-report/vital-statistics-on-congress/ 728 Dugan and Chenoweth We next compare the yearly trends for different measures of violent hate crime rates against Black and Latinx persons to assess which rates better measure the underlying pattern of violent hate crimes in the United States. We then compare the yearly trends of the two types of govern- ment actions with their corresponding violent hate crime rate. In this section, we also summarize all key variables to show the average level of each over the entire period. To test our hypotheses, we combine the monthly data from each source that are specific to each protected group to run two distinct models. Monthly units are used to capture close to real- time reactions to government action, while reducing sparseness in the data. Close to real-time reactionsarepreferredashatecrimeoffendersareoftenimpulsive(Levin&McDevitt,2002).Also, King and Sutton (2013), who relied on daily time series of hate crime events following antecedent events, discovered that hate crime spikes a few days after each event and then decays over the next several days. As the current research assesses the effects of cumulative lower level events that we characterize as positive or negative attention, as opposed to single antecedent events (e.g., the O.J. Simpson verdict), we anticipate that their effects on hate crimes will still be detectable during the following month as attention is ongoing. Furthermore, we recognize that government actions could also be influenced by preceding violent hate crimes, which could, in turn, bias our findings through simultaneity. In fact, the GATE-USA data show that after the burning of several Black churches, President Clinton began a national campaign to improve race relations, demonstrating that violent hate crimes affect gov- ernment actions relevant to the victims of those crimes. Therefore, to assure that the specified results estimate the effects of government actions on violent hate crimes, and are not biased as a result of this type of endogeneity, we conduct three-equation reduced form vector autoregression (VAR) analyses with exogenous control variables (shown in equation 1) for Black and Latinx per- sons using monthly data. We see that in these series of equations, VAR simultaneously estimates the effects of lagged pro-group government actions (PGAs) and lagged anti-group government actions (AGAs) on the violent hate crime rate against the specified group (VHCR) while estimat- ing the effects of the same series of lagged variables on each type of government action. In these series, X includes the control variables listed in table 1, the number of civil rights groups, indi- cators of whether the federal government was controlled by the Democrats or Republicans, the violent crime rate, the total number of participating agencies that quarter, the percentage of par- ticipating agencies that report at least one incident that quarter, and indicators for each month, excluding December. Granger causality tests are conducted to assess the directionality of the rela- tionship, and impulse response functions (IRFs) are generated to show how the VHCR responds to the impulses of PGA and AGA, as well as how PGA and AGA respond to the impulse of the VHCR. 𝑉𝐻𝐶𝑅𝑡 = 𝐽∑ 𝑗=1 𝛽1𝑗𝑉𝐻𝐶𝑅𝑡−𝑗 + 𝐽∑ 𝑗=1 𝛽2𝑗𝑃𝐺𝐴𝑡−𝑗 + 𝐽∑ 𝑗=1 𝛽3𝑗𝐴𝐺𝐴𝑡−𝑗 + 𝛃𝟒𝐗 𝑃𝐺𝐴𝑡 = 𝐽∑ 𝑗=1 𝛼1𝑗𝑃𝐺𝐴𝑡−𝑗 + 𝐽∑ 𝑗=1 𝛼2𝑗𝑉𝐻𝐶𝑅𝑡−𝑗 + 𝐽∑ 𝑗=1 𝛼3𝑗𝐴𝐺𝐴𝑡−𝑗 + 𝛂𝟒𝐗 𝐴𝐺𝐴𝑡 = 𝐽∑ 𝑗=1 𝛾1𝑗𝐴𝐺𝐴𝑡−𝑗 + 𝐽∑ 𝑗=1 𝛾2𝑗𝑉𝐻𝐶𝑅𝑡−𝑗 + 𝐽∑ 𝑗=1 𝛾3𝑗𝑃𝐺𝐴𝑡−𝑗 + 𝛄𝟒𝐗 (1) Dugan and Chenoweth 729 BecausetheVHCRsweregeneratedfromachangingmixtureofagenciesthatparticipatedinthe HCSA, we replicate the analyses shown in equation 1 using only those agencies that participated in all four quarters for all 21 years and for at least 20 of the 21 years. Additional sensitivity analyses are conducted and reported in detail in appendix D in the online supporting information to further assess whether the findings hold if we use the entire U.S. popu- lation in the denominator instead of only counting those persons represented by the participating agencies. Also, to better validate the HCSA, we construct total hate crime data from the HCSA and the NCVS and compare the rates from the HCSA with those in the NCVS that were known and unknown to the police in appendix E in the online supporting information. We also run the VAR models shown in equation 1 and compare the IRFs across all three models. We expect that the results generated from the HCSA will be more similar to the NCVS police known hate crimes than to the police unknown hate crimes. 5 RESULTS Before turning to our analysis, we assess how well the distribution of participating agencies rep- resents the entire nation. Detailed analyses are found in appendix B in the online supporting information. The UCR HCSA files used here include information from 22,213 law enforcement agencies from 1992 to 2012. Only 3.3 percent of those agencies (743) participated (i.e., submitted a report) for every quarter during the 21-year period. A larger subset, 8.9 percent or 1,972 agencies, participated in 20 of the 21 years. The states that have the highest average number of participating years per agency are Texas (17.2), Iowa (16.2), and New York (15.9), whereas agencies in Hawaii (0), Alaska (.47), and Georgia (1.95) have the lowest participation. The rates generated from the subsets of participating agencies are dominated by Texas agencies, which account for 36 percent of those that participated for at least 20 years and 66 percent of those that participated for all 21. The main analyses are replicated below using violent hate crime rates generated from these two data sets to assess the robustness of the findings. Figure 1 presents the annual trends of violent hate crime rates against Black (a) and Latinx (b) persons with both the total population and the participating agency population used in the denominators. In each case, the rate using the limited population is larger by a factor that peaks at 2.66 in 1992 and is as small as 1.02 in 2010 for both groups. The dashed line in both cases shows a nearly monotonic decreasing trend from 1992 to 2012. Yet, when we look at the rates that use the entire population in the denominator, we see that they both increase in the beginning of the series, peak in 1996, and then follow a similar trend as the one that uses the limited population in the denominator. The divergence in the beginning of the series mimics what would be expected if agency compliance started low and then grew. The convergence later in the series suggests more of an equilibrium, indicating that the findings generated by each rate might converge. All forth- coming analyses use the rates generated from the agency populations in the denominators. Figures 2a and 2b shows the yearly trends for the specific types of government actions and their corresponding violent hate crime (VHCR) rate per million using the agency populations as denominators. The VHCR is measured on the left axis, and the total number of monthly gov- ernment actions is measured on the right axis. A comparison of government actions shows that that U.S. federal government acted more frequently in support of and against Latinx persons or immigration compared with when African American issues were addressed. More often than not, when federal officials directly addressed issues relevant to Black persons during this period, they did so in a way that supported them. The left axes show that the peak in the VHCR against 730 Dugan and Chenoweth FIGURE 1 Violent hate crime rate per protected group using different populations in the denominator (n = 21) Dugan and Chenoweth 731 FIGURE 2 Government actions and VHCRs relevant to each group (n = 21) 732 Dugan and Chenoweth TABLE 2 Descriptive statistics for key variables (n = 252) Variables Mean St. Dev Min Max p (zero) Anti-Black VHCR .735 .324 .267 .122 .000 Anti-Latinx VHCR .144 .053 .035 .328 .000 Pro-Black GA .460 .898 .000 6.000 .702 Anti-Black GA .127 .480 .000 6.000 .893 Pro-Latinx/Immigrant GA .425 .787 .000 5.000 .694 Anti-Latinx/Immigrant GA .448 .758 .000 4.000 .663 Black people is nearly six times higher than that for Latinx persons. In general, the trends show a greater number of relevant federal actions in the earlier years, which seems to correspond to the downward trends of the VHCR. The first 2 to 3 years of the series, however, show lower numbers of government actions. Next, we summarize the number of government actions and violent hate crimes relevant to Black and Latinx persons. 5.1 Summary statistics Table 2 lists the descriptive statistics for the key dependent and independent variables. Turning firsttothedependentvariables, theaverageVHCR ishighestagainstBlackpeople(.74permillion), whereas that for Latinx persons averages about .14 per million. The average number of government actions (GA) relevant to these groups is less than one per month. Furthermore, the government more often gives positive attention to these groups than negative attention. We see that most actions are relevant to Latinx persons or immigra- tion, with nearly the same average number of actions each month (.43 favor and .45 against). On average there were .46 favorable federal actions per month for African Americans (or one every 2.17 months), and adverse actions once every 7.7 months. The last column shows the proportion of months where no government actions were reported for each protected group. Here we see that in most months, members of the U.S. federal government engage in no actions against these groups. Furthermore, federal officials expressed or implemented favorable actions toward Black and Latinx persons or immigrants in about 30 percent of the months. 5.2 Vector autoregression models Dickey-Fuller tests that allow four lags find that all six variables used in the VAR regression are stationary.Thetest for theVHCRforBlackpersons,however, isonlysignificantat the .10level.9,10 The optimal number of lag lengths is selected by running the VAR models with different lag lengths and comparing the likelihood ratio test statistic, the Akaike’s information criterion, and 9 Because of this and the downward annual trends shown in figure 2, additional tests were run on the two VHCRs. A modified Dickey-Fuller test that applies a GLS transformation shows that both series are stationary in early lags but non- stationary with later lags. The Phillips-Perron test concludes that both are stationary. 10 As a test for robustness, the primary VAR model was rerun using the first differenced anti-Black VHCR; the results remained substantively similar. Dugan and Chenoweth 733 TABLE 3 Granger causality wald tests (n = 250) Black Latinx VHCR Pro-Group Action 21.313** .871 Anti-Group Action 7.112** 5.980* Pro-Group Action VHCR 3.077 2.955 Anti-Group Action 4.056 .968 Anti-Group Action VHCR 1.433 .330 Pro-Group Action 4.191 4.376 *p ≤ .05; **p ≤ .01. theHannan and Quinn’s information criterion. For both groups, the two-lagged model is selected. Tests on the residuals show no autocorrelation, confirming the adequacy of the choice of the two- lagged model for each group. Finally, all eigenvalues generated from the companion matrices fall within the unit root, allowing us to conclude that both VAR models are stationary. Tests for nor- mality show that both VHCRs are normally distributed but the government actions are not. This is unsurprising, given the high proportion of zeros and the square-rooted counts. Findings that use the government actions as the response variable will be interpreted with caution. Their results, however, do correspond to unpublished structural equation models of a similar specification that were estimated using the negative binomial distribution. Table 3 reports the Wald chi-square statistics for tests of Granger causality in both VAR models. Here we find that only the hypothesized relationships show any evidence of Granger causality. Government actions seem to influence the VHCR, but the VHCR does not seem to influence gov- ernment actions. The tests indicate that both pro- and anti-African-American actions relate to the VHCR against Black people, whereas only anti-immigration/Latinx actions relate to the VHCR against Latinx persons. When we consider these results within the context of the hypotheses, Black people might be vulnerable to both political threat and emboldenment and Latinx persons might only be vulnerable to emboldenment. As the Granger causality tests cannot speak to direc- tionality, we turn now to the IRFs to more specifically assess the hypotheses. Figures 3 and 4 show the IRFs (and their 95 percent confidence bounds) of anti- and pro- group government actions as the impulse on VHCR responses (a) and the inverse (b; VHCR as the impulse on government actions responses) for Black and Latinx persons, respectively. In figure 3a, we see that only those acts that give positive attention to African Americans appear to influence the VHCR, despite both types of actions showing Granger causality in table 3. This supports the political threat hypothesis but not the emboldenment hypothesis. The IRF peaks at .03 at the first month after a 1 standard deviation shock in pro-African American actions. It then drops to insignificance in the second month, indicating only a short-term effect. Figure 3b shows that the VHCR against Black people is positively related to pro-African American actions by the federal government, suggesting a simultaneous relationship, despite the Granger results reported above. Here we see that the effect is only significant in the same month as the impulse with a magnitude of .08. Forecast error variance decompositions (FEVD) show that pro-African Americangovernmentactionsaccountforapproximately 8.4percentof thevariation in the VHCR against Black people. 734 Dugan and Chenoweth FIGURE 3 Impulse Response Functions for Black VAR (n = 250) [Color figure can be viewed at wileyon- linelibrary.com] a). Action as Impulse on Violent Hate Crime Response b). Violent Hate Crime as Impulse on Action Response Dugan and Chenoweth 735 FIGURE 4 Impulse Response Functions for Latinx VAR (n = 250) [Color figure can be viewed at wileyon- linelibrary.com] a). Action as Impulse on Violent Hate Crime Response b). Violent Hate Crime as Impulse on Action Response 736 Dugan and Chenoweth TABLE 4 VAR coefficients and standard errors of exogenous variables estimating effects on violent hate crime rates for Black and Latinx persons (n = 250) Black Latinx Variables Coef SE Coef SE Civil Rights Groups −.120** (.040) −.101 (.073) Democratic Controlled −.047** (.018) −.011 (.038) Republican Controlled −.046** (.018) −.101** (.037) Violent Crime Rate .134** (.026) .206** (.051) Total Participating Agencies −.020** (.008) .023 (.016) Percent Agencies w/Incids .032** (.010) .107** (.021) January .340** (.031) .347** (.065) February .379** (.039) .276** (.074) March .458** (.032) .472** (.063) April .391** (.036) .429** (.069) May .358** (.032) .442** (.065) June .324** (.034) .394** (.067) July .322** (.032) .315** (.066) August .352** (.033) .332** (.064) September .370** (.033) .442** (.065) October .356** (.034) .372** (.066) November .163** (.037) .234 (.072) *p ≤ .05; **p ≤ .01 (two-tailed). TheIRFsshowninfigure4asuggestthatLatinxpersonsarevulnerabletoemboldenmentrather thantopoliticalthreat,astheanti-immigrationfunctionspikesintheleftgraph,whereastheother never clears zero. Furthermore, we find no evidence of simultaneity in figure 4b. The magnitude of a 1 standard deviation shock of anti-Latinx government actions also peaks at .03 in the month following the shock, and then it falls to insignificance, suggesting only a short-term effect. FEVD shows that this only accounts for 2.1 percent of the variance of the VHCR against Latinx persons. Table 4 presents the coefficient estimates of the control variables on the VHCR for each group. The coefficient estimates of the control variables on the two types of government actions are reported in appendix C in the online supporting information. Here we find that the number of civil rights groups is negatively, not positively, related to violent hate crimes for Black people. This is surprising because both political threat and research by McVeigh et al. (2003) predict that more civil rights groups would either lead to more violence (political threat) or more reporting of hate crimes (advocacy). The negative effects suggest that the growth in civil rights groups are related to fewer violent hate crimes perhaps because they are effectively changing attitudes or prevent- ing harmful altercations. Although the difference between this finding and McVeigh et al.’s is somewhat puzzling, the sources of variation in the two estimates differ substantially. Their anal- ysis relied on cross-county variation within 1 year (2000), and the variation that contributed to this estimate is the yearly changes in civil rights groups over 21 years. Furthermore, this estimate might have been influenced by the changes in agency participation in the HCSA over time. Given that the different conclusions are based on different sources of the variation and that the current estimate is based on only 21 unique values over the entire series, more research is needed to assess the impact of civil rights groups on the reporting of hate crimes. Dugan and Chenoweth 737 When the federal government was entirely controlled by Democrats, violent hate crimes against Black people were lower; and when Republicans were in control, violent hate crimes against both Black and Latinx persons were lower. This suggests that motivations to attack might be muted when the government is operating more smoothly, which is aligned with Young and Dugan’s (2011) argument that when governments have fewer veto players, domestic terrorists perpetrate fewer attacks. Unsurprisingly, the violent hate crime rate is positively and significantly associated with the VHCR against Black and Latinx persons. The total number of participating agencies is included because we would expect that the more agencies that submit reports to the HCSA program, the more violent hate crimes will be reported; however, this measure is negatively related to violence against Black people and unrelated to violence against Latinx persons. As expected, we see that the percentage of participating agencies that submitted an incident report is positively related to violent hate crimes against both groups. Finally, all months are statistically significant and positive, suggesting that violent hate crimes are less common in December than in other months. Insum,examiningdifferentpatternsinhatecrimesagainstmarginalizedgroups,wefindstrong supportforpoliticalthreattheoryforanti-Blackhatecrimes.Wefindnosupportforpoliticalthreat theory for Latinx persons, however, and instead find evidence of emboldenment. To assess the robustness of these findings to the distribution of participating agencies, we replicate the analysis using hate crime rates generated from only those agencies that participated consistently. First we compare the trends across samples. Figure 5 shows the VHCR rates against (a) Black people and (b) Latinx persons across all three sets of participating agencies. The trends appear remarkably close, especially for the anti-Black VHCRs. The monthly trend correlations between all agencies and the 20+-year agencies are .94 for anti-Black VHCRs and .79 for anti-Latinx VHCRs; and for the 21-year agencies, .89 and .44, respectively. The corresponding first differenced correlations are .65 and .38 for the anti-Black VHCRs and .58 and .35 for the anti-Latinx VHCRs. The IRFs showing the impulse of anti- and pro-group actions on the VHCR response generated from the two subsets of participating agencies are presented in figure 6. Figures 6a and 6b show that the support for the political threat hypothesis in the African American model is robust to the composition of the participating agencies as the pro-Black IRF and its confidence bounds clear the zero line in both models with peaks that are slightly higher than that in the original model (.036 and .037). Figures 6c and 6d show that the emboldenment hypothesis in the Latinx model is also still supported in these models, with the confidence bounds of the IRF clearing zero at the peak in both models (.05 and .10). Additional tests of robustness are found in appendices D and E in the online supporting infor- mation. Appendix D shows that when we use the total U.S. population to generate the VHCR, the results are nearly identical. Appendix E compares the total hate crime rates generated by the UCR HCSAdataandtheNationalCrimeVictimizationSurvey(NCVS)data. Inthoseanalyses,wecom- pare the UCR hate crimes with the NCVS hate crimes known to the police and unknown to the police. The UCR rates are more strongly correlated (both linearly and first differenced) with the rate of police known hate crime events generated by the NCVS than the rate of events unknown to police, although differences remain. Furthermore, a comparison of main results from all three sources shows that the IFRs generated from the HCSA data look more similar to those generated from the police known NCVS data than those unknown to the police. The appendix describes in detail the differences in the two data sources, concluding that the comparison provides some sup- port that the data are measuring the same underlying dynamics. It is only when we consider these 738 Dugan and Chenoweth FIGURE 5 VHCR generated from different sets of participating agencies (n = 21) Dugan and Chenoweth 739 FIGURE 6 Impulse Response Functions for Black and Latinx persons using rates from subsets of participat- ing agencies (n = 250) [Color figure can be viewed at wileyonlinelibrary.com] a). Black IRFs using Agencies With at Least 20 Years of Participation b). Black IRFs using Agencies With all 21 Years of Participation c). Latinx IRFs using Agencies With at Least 20 Years of Participation d). Latinx IRFs using Agencies With all 21 Years of Participation 740 Dugan and Chenoweth FIGURE 6 Continued Dugan and Chenoweth 741 results and the other robustness tests, however, that we feel confident that we are estimating the appropriate relationships. 6 DISCUSSION This article began as a search for evidence about whether changes in the political power of dif- ferent marginalized groups in the United States affect the amount of hate-motivated violence directed toward them. We were drawn to this research by the reported rise in hate-based violence against minority groups that drew hostile attacks from the Trump administration. We developed two compatible hypothesespredicting increasedviolence toward minority groups after the federal government takes either favorable or unfavorable action toward these groups. Political threat the- ory concludes that minority groups are at risk of backlash after expanding their political power, as dominant groups use violence to reestablish equilibrium. Indeed, we found an increase in vio- lent anti-Black hate crimes after the U.S. government took actions that were favorable toward African Americans. We extended this theory to account for surges in dominant group resources when political elites openly express hostility toward minority groups. Support for the embolden- ment hypothesis was apparent only in our findings for Latinx persons. As the federal government opposed immigrants generally, or Latinx persons specifically, violent hate crimes against Latinx persons increased. The data used to generate these findings suffer notable limitations as surely some U.S. actions and violent hate crimes are undetected during collection. We feel confident, however, that the GATE data captured the larger patterns of attention given to these groups as the description pro- vided in appendix A in the online supporting information characterizes most of the major events over this period. Also, the GATE data capture more of the subtle attention these groups receive as political leaders promote certain initiatives and Congress considers legislative action. Further- more, the variation used to generate our estimates was based on month-to-month changes, which are less sensitive to undercounting of both sources. Despite this, we recognize that both anti- and pro-group government actions are measured with error, and that this type of measurement error is likely correlated, possibly causing estimation problems. Supplemental analysis not shown here, however, produce the same findings when the VAR models are run with only one type of government action. We also recognize that the estimated positive effects for African Ameri- cans could also be driven by an increased reporting by victims in those groups when the federal government expressly supports their movement. Yet, this bias might also be counterbalanced by reluctant reporting agencies that react to positive attention by dismissing hate crimes as neutrally motivated. The findings hold in our supplemental analyses, when we use subsets of agencies that partici- patedintheHCSAprogramforallornearlyallyearsinouranalysis.Furthermore,acomparisonof hate crime rates generated from both the HCSA and NCVS shows that the HCSA data align more closelyto victimization eventsknown to thepolice thanthose unknownto thepolice, which helps usfeelmoreconfidentthattheHCSAiscapturingtheunderlyingdynamicofhatecrimesreported to the police. Despite the possibility that Black people might be more likely to report hate crimes when the federal government is exhibiting support for civil rights, we find it highly unlikely that Latinx victims of violent hate crime would increase their reporting to police when the political climate is hostile toward immigrants, as research suggests that some Latinx persons in the U.S. view contact with police as risky (Vidales, Day, & Powe, 2009; Xie & Baumer, 2019). While the fed- eral hate crime statistics are far from perfect, the fact that our models—which were derived from 742 Dugan and Chenoweth data combined from two much different collection efforts—are consistent raises confidence that the hate crime data were sufficient for our purposes. Combined, we find support for the political threat hypothesis among hate crimes committed against Black people and for the emboldenment hypothesis among hate crimes committed against Latinx persons. We would be remiss, however, to suggest that these findings can be generalized to all juris- dictions in the United States. The voluntary participation, and differential efforts to investigate crimes for hateful biases undeniably exclude important jurisdictions, which results in selection bias. This raises the question of how selection might affect the main results. If agencies in more biased jurisdictions are reluctant to participate, as suggested in the findings by McVeigh et al. (2003)andKing(2007), thentheeffectsfromonlytheparticipatingagenciesmightbesmallerthan those that would be estimated had all agencies fully participated. In other words, following King’s logic, if those jurisdictions that are more “threatened” by minorities are less likely to participate, then the current set of results that rely exclusively on participating agencies are surely muted. In fact, when we reanalyze the model using data only from agencies that fully participated in the populous for the increasingly racially diverse state of Texas, the African American political threat effect increases by a factor of four (peak = .12). The Latinx emboldenment effect also increases by the same magnitude when we use data from all participating Texas agencies. If jurisdictions in Texas look more like those without agency participation than those with agency participation, then these comparisons do, indeed, suggest that the bias in the primary analyses is downward. The dependency of the findings on the composition of participating agencies reminds us that hatecrimesarelocalphenomena,andthat if theyareresponsivetoactionsbyfederal leaders, then they are likely sensitive to those by nearby officials. Continued research in this area should incor- poratevariationinstateandlocalgovernmentactions, includingeffortstoholdpoliceaccountable for violence against Black people and to remove undocumented individuals from being eligible for statewide benefits. Taken together, what do our findings suggest about the reported increase in hate crimes since Donald Trump was elected as president? Although it is an out-of-sample case, we argue that the rise in hate crimes against Latinx people since the 2016 election appears to be a continuation of patterns that have persisted over recent decades. Since the 1990s, federal government actions targeting immigrants and Latinx persons seemed to have emboldened violent hate crimes against them. Moreover, federal efforts to support Black people are complicated by their risk of backlash, as supportive federal rhetoric, actions, and policies tend to trigger violent anti-Black hate crimes. Theseeffects likely reflect long-simmeringinstitutionalized andinternalizedracism in theUnited States, which is triggered when federal officials send positive cues about this demographic. These mixed findings point to the need to disaggregate the implications of political threat and emboldenment hypotheses by the affected groups, rather than assuming these mechanisms operate identically across groups. Although our data are insufficient to identify specific mecha- nisms that could be driving the differences across groups, one possibility is that White perpetra- tors stereotype Black Americans as “insiders” and Latinx people as “outsiders.” It may be that members of the dominant group commit violence against perceived insiders to punish them for increases in power, whereas members of the dominant group commit violence against perceived outsiders when the federal government signals that the threat from outsiders is growing. Quali- tative data that document the reasoning of hate crime perpetrators could provide insight into the nature of perceived threat or other emotions that leads to this type of violence. Another limita- tion that qualitative data might uncover would be the effects, if any, federal government actions have on people who possess multiple overlapping identities, such as African Americans of Latinx heritage, or whether perpetrators of hate crime substitute one targeted group for another. Dugan and Chenoweth 743 In future research, scholars should replicate this analysis with more recent data to directly assess the degree to which Trump’s election and subsequent presidency has triggered an embold- ening from the far-right. In fact, the daily documentation of the President’s Tweets provide a rich source of information regarding the rhetoric used toward different marginalized groups. Addi- tionally, future research should assess the role of emboldenment on mobilizing acts by organized White supremacist groups. That said, this research, which draws on two decades of U.S. government activity, suggests that federalactions matter in important ways that could motivate violent hate crimes against protected groups—either by supporting them without protecting them from possible backlash or by criti- cizing and undermining them, emboldening perpetrators to commit hate crimes against them. Indeed, our results confirm evidence from other studies suggesting that hate speech among elites canmotivatehatecrimesamongconstituents(Gagnon,1995;Müller&Schwarz,2018).Thesefind- ings have important implications for preventing hate crimes, as well as for assigning culpability when they occur. ORCID LauraDugan https://orcid.org/0000-0001-5930-6950 EricaChenoweth https://orcid.org/0000-0003-3113-2439 REFERENCES Alexander, M. (2012). The new Jim Crow: Mass incarceration in the age of colorblindness. New Press. Baumer,E.P.,Messner,S.F.,&Rosenfeld,R.(2003).Explainingspatialvariationinsupportforcapitalpunishment: A multilevel analysis. American Journal of Sociology, 108, 844–875. https://doi.org/10.1086/367921 Bell, J. (2002). Policing hatred: Law enforcement, civil rights, and hate crime. New York University Press. Behrens, A., Uggen, C., & Manza, J. (2003). Ballot manipulation and the “menace of negro domination”: Racial threat and felon disenfranchisement in the United States, 1850–2002. American Journal of Sociology, 109, 559– 605. https://doi.org/10.1086/378647 Blalock, H. M. Jr (1967). Toward a theory of minority-group relations. Wiley. Blumer, H. (1958). Race prejudice as a sense of group position. The Pacific Sociological Review, 1, 3–7. https://doi. org/10.2307/1388607 Bouie, J. (2017, June 1). What we have unleashed. This year’s string of brutal hate crimes is intrinsically connected to the rise of Trump. Slate, 1. Boyd, E. A., Berk, R. A., & Hamner, K. M. (1996). “Motivated by hatred or prejudice”: Categorization of hate- motivated crimes in two police divisions. Law & Society Review, 30, 819–850. https://doi.org/10.2307/3054119 Brown, M. C., & Warner, B. D. (1992). Immigrants, urban politics, and policing in 1900. American Sociological Review, 57, 293–305. https://doi.org/10.2307/2096236 Brown, v. Board of Education of Topeka (1954). 347 U.S. 483. Burch, A. D. S. (2017, April 30). Spread of hate crimes has lawmakers seeking harsher penalties. New York: Times, Retrieved from https://www.nytimes.com/2017/04/30/us/hate-crimes-legislation.html Charney, C. R. (1999). Civil society, political violence, and democratic transitions: Business and the peace process in South Africa, 1990 to 1994. Comparative Studies in Society and History, 41, 182–206. https://doi.org/10.1017/ S0010417599001917 Cronin, S. W., McDevitt, J., Farrell, A., & Nolan III, J. J. (2007). Bias-crime reporting organizational responses to ambiguity, uncertainty, and infrequency in eight police departments. American Behavioral Scientist, 51, 213–213. https://doi.org/10.1177/0002764207306054 Cunningham, D. (2012). Mobilizing ethnic competition. Theory and Sociology, 41, 505–525. https://doi.org/10.1007/ s11186-012-9178-4 Doane, A. W. (1997). Dominant group ethnic identity in the United States: The role of “hidden” ethnicity in inter- group relations. The Sociological Quarterly, 38, 375–397. https://doi.org/10.1111/j.1533-8525.1997.tb00483.x https://orcid.org/0000-0001-5930-6950 https://orcid.org/0000-0001-5930-6950 https://orcid.org/0000-0003-3113-2439 https://orcid.org/0000-0003-3113-2439 https://doi.org/10.1086/367921 https://doi.org/10.1086/378647 https://doi.org/10.2307/1388607 https://doi.org/10.2307/1388607 https://doi.org/10.2307/3054119 https://doi.org/10.2307/2096236 https://www.nytimes.com/2017/04/30/us/hate-crimes-legislation.html https://doi.org/10.1017/S0010417599001917 https://doi.org/10.1017/S0010417599001917 https://doi.org/10.1177/0002764207306054 https://doi.org/10.1007/s11186-012-9178-4 https://doi.org/10.1007/s11186-012-9178-4 https://doi.org/10.1111/j.1533-8525.1997.tb00483.x 744 Dugan and Chenoweth Dugan, L., & Chenoweth, E. (2017, June 29). Introducing government actions in terror environments – United States (GATE-USA) [Final Report]. START. Dugan, L., Huang, J., LaFree, G., & McCauley, C. (2008). Sudden desistance from terrorism: The Armenian Secret Army for the Liberation of Armenia and the Justice Commandos of the Armenian Genocide. Dynamicsof Asym- metric Conflict, 1, 231–249. https://doi.org/10.1080/17467580902838227 Eitle, D., D’Alessio, S. J., & Stolzenberg, L. (2002). Racial threat and social control: A test of political, economic, and threat of black crime hypotheses. Social Forces, 81, 557–576. https://doi.org/10.1353/sof.2003.0007 Ellsworth, S. (1992). Death in a promised land: The Tulsa race riot of 1921. Louisiana State University Press. Epps,C.R.(1998).Therightsrevolutions:Lawyers,activists,andsupremecourtsincomparativeperspective.University of Chicago Press. Fobanjong, J. (2003). Understanding the backlash against Affirmative Action. Nova Science Press. Fossett, M. A., & Kiecolt, K. J. (1989). The relative size of minority populations and white racial attitudes. Social Science Quarterly, 70, 820–835. Fourie, A., & Eloff, T. (2005). The case for collective business action to achieve systems change: Exploring the contributions made by the private sector to the social, economic and political transformation process in South Africa. Journal of Corporate Citizenship, 18, 39–48. Freilich, J. D., Chermak, S. M., Belli, R., Gruenewald, J., & Parkin, W. S. (2014). Introducing the United States Extremist Crime Database (ECDB). Terrorism and Political Violence, 26, 372–384. https://doi.org/10.1080/ 09546553.2012.713229 Gagnon Jr., V. P. (1995). Ethnic nationalism and international conflict: The case of Serbia. International Security, 19, 130–166. Giles, M. W., & Evans, A. (1986). The power approach to intergroup hostility. Journal of Conflict Resolution, 30, 469–486. https://doi.org/10.1177/0022002786030003004 Giles, M. W., & Hertz, K. (1994). Racial threat and partisan identification. The American Political Science Review, 88, 317–326. https://doi.org/10.2307/2944706 Goldman, H. (2017, February 24). NYPD links hate crimes increase to 2016 presidential campaign. Bloomberg News. Retrieved from https://www.bloomberg.com/news/articles/2017-02-24/nypd-links-hate-crimes-increase-to- 2016-presidential-campaign Green, D. P., Strolovitch, D. Z., & Wong, J. S. (1998). Defended neighborhoods, integration, and racially motivated crime. American Journal of Sociology, 104, 372–403. https://doi.org/10.1086/210042 Hart, T., & Rennison, C. (2003, March). Reporting crime to the police, 1992–2000 (Bureau of Justice Statistics Special Report). Hauslohner, A. (2018, May 11). Hate crimes jump for fourth straight year in largest U.S. cities, study shows. The Washington Post Retrieved from https://www.washingtonpost.com/news/post-nation/wp/2018/05/11/hate- crime-rates-are-still-on-the-rise/?utm_term=.938d1953cd0b Jackson, P. I., & Carroll, L. (1981). Race and the war on crime: The sociopolitical determinants of municipal police expenditures in 90 non-southern U.S. cities. American Sociological Review, 46, 290–305. https://doi.org/10.2307/ 2095061 Jacobs, D., & Carmichael, J. T. (2001). The politics of punishment across time and space: A pooled time-series analysis of imprisonment rates, Social Forces, 80, 61–89. https://doi.org/10.1353/sof.2001.0070 Jacobs, D., & Helms, R. (1999). Collective outbursts, politics, and punitive resources: Toward a political sociology of spending on social control. Social Forces, 77, 1497–1523. https://doi.org/10.1093/sf/77.4.1497 Jenness, V., & Grattet, R. (2001). Making hate a crime: From social movement to law enforcement. Russell: Sage Foundation. Kalmoe, N. P. (2014). Fueling the fire: Violent metaphors, trait aggression, and support for political violence. Polit- ical Communication, 31, 545–563. https://doi.org/10.1080/10584609.2013.852642 King, R. D. (2007). The context of minority group threat: Race, institutions, and complying with hate crime law. Law & Society Review, 41, 189–224. https://doi.org/10.1111/j.1540-5893.2007.00295.x King,R.D.,&Sutton,G.M.(2013).Hightimesforhatecrimes:Explainingthetemporalclusteringofhate-motivated offending. Criminology, 51, 871–894. https://doi.org/10.1111/1745-9125.12022 King, R. D., & Wheelock, D. (2007). Group threat and social control: Race, perceptions of minorities and the desire to punish. Social Forces, 85, 1255–1280. https://doi.org/10.1353/sof.2007.0045 https://doi.org/10.1080/17467580902838227 https://doi.org/10.1353/sof.2003.0007 https://doi.org/10.1080/09546553.2012.713229 https://doi.org/10.1080/09546553.2012.713229 https://doi.org/10.1177/0022002786030003004 https://doi.org/10.2307/2944706 https://www.bloomberg.com/news/articles/2017-02-24/nypd-links-hate-crimes-increase-to-2016-presidential-campaign https://www.bloomberg.com/news/articles/2017-02-24/nypd-links-hate-crimes-increase-to-2016-presidential-campaign https://doi.org/10.1086/210042 https://www.washingtonpost.com/news/post-nation/wp/2018/05/11/hate-crime-rates-are-still-on-the-rise/?utm_term=.938d1953cd0b https://www.washingtonpost.com/news/post-nation/wp/2018/05/11/hate-crime-rates-are-still-on-the-rise/?utm_term=.938d1953cd0b https://doi.org/10.2307/2095061 https://doi.org/10.2307/2095061 https://doi.org/10.1353/sof.2001.0070 https://doi.org/10.1093/sf/77.4.1497 https://doi.org/10.1080/10584609.2013.852642 https://doi.org/10.1111/j.1540-5893.2007.00295.x https://doi.org/10.1111/1745-9125.12022 https://doi.org/10.1353/sof.2007.0045 Dugan and Chenoweth 745 Kteily, N., Bruneau, E., Waytz, A., & Cotterill, S. (2015). The ascent of man: Theoretical and empirical evidence for blatant dehumanization. Journal of Personality and Social Psychology, 109, 901–931. https://doi.org/10.1037/ pspp0000048 Levin, B. (2002). From slavery to hate crime laws: The emergence of race and status-based protection in American criminal law. Journal of Social Issues, 58, 227–245. https://doi.org/10.1111/1540-4560.00258 Levin, B., & Reitzel, J. D. (2018). Hate crimes rise in U.S. cities and counties in time of division & foreign interference (Report by the Center for the Study of Hate & Extremism). California State University, San Bernadine. Levin, J., & McDevitt, J. (2003). Hate crimes revisited: America’s war on those who are different. Westview Press. Lipton, S. (2015, December 13). “The words that killed medieval Jews.” New York Times. Retrieved from https: //www.nytimes.com/2015/12/13/opinion/the-words-that-killed-medieval-jews.html Lopez, M. H., Morin, R., & Taylor, P. (2010, October 28). Illegal immigration backlash worries, divides Latinos. Pew Research Center. Retrieved from http://www.pewhispanic.org/2010/10/28/illegal-immigration-backlash- worries-divides-latinos/ Martin, S. E. (1995). “A cross-burning is not just an arson”: Police social construction of hate crimes in Baltimore County. Criminology, 33, 303–326. https://doi.org/10.1111/j.1745-9125.1995.tb01180.x McVeigh, R., Cunningham, D., & Farrell, J. (2014). Political polarization as a social movement outcome: 1960s Klan activism and its enduring impact on political realignment in southern counties, 1960 to 2000. American Sociological Review, 79, 1144–1171. https://doi.org/10.1177/0003122414555885 McVeigh, R., Welch, M. R., & Bjarnason, T. (2003). Hate crime reporting as a successful social movement outcome. American Sociological Review, 68, 843–867. https://doi.org/10.2307/1519748 Müller, K., & Schwarz, C. (2018). Making America hate again? Twitter and hate crime under Trump (Working paper). University of Warwick. SSRN: Retrieved from https://ssrn.com/abstract=3149103 or https://doi.org/10. 2139/ssrn.3149103 NCCS.(n.d.).GuidetousingNCCSdata.NationalCenterforCharitableStatistics.Retrievedfromhttp://nccs.urban. org/ OfficeofJusticePrograms.(2018,April14–18).Hatecrime.2018NationalCrimeVictims’RightsWeekResourceGuide: Crime and Victimization Fact Sheets. Retrieved from https://ovc.ncjrs.gov/ncvrw2018/fact_sheet.html OklahomaCommissiontoStudytheRaceRiotof1921. (2001,February28).Tulsaraceriot:AreportbytheOklahoma Commission to Study the Race Riot of 1921. Pew Research Center. (2018, June 28). Shifting public views on legal immigration into the U.S. Phillips, A. (2017, June 16). “They’re rapists.” President Trump’s campaign launch speech two years later, anno- tated. Washington Post, Retrieved from https://www.washingtonpost.com/news/the-fix/wp/2017/06/16/theyre- rapists-presidents-trump-campaign-launch-speech-two-years-later-annotated/?utm_term=.e279b17a93b8 Potok, M. (2017). Hate and extremism. The radical right was more successful in entering the political mainstream last year than in half a century. How did it happen?” Intelligence Report, by the Southern Poverty Law Center, 162, Spring). Quillian, L. (1995). Prejudice as a response to perceived group threat: Population composition and anti-immigrant and racial prejudice in Europe. American Sociological Review, 60, 586–611. https://doi.org/10.2307/2096296 Quillian, L. (1996). Group threat and regional change in attitudes toward African-Americans. American Journal of Sociology, 102, 816–860. https://doi.org/10.1086/230998 Schrodt, P. (2012). CAMEO: Conflict and mediation event observations event and actor codebook. Computational Event Data Systems. Smith, B. L., & Damphousse, K. R. (2001). American terrorism study: Patterns of behavior, investigation and prose- cution of American terrorists (Final Report). National Institute of Justice, NIJ Grant # 1999-IJCX-0005. START. (2016). Global Terrorism Database [Data file]. Taylor, K-Y. (2016). From #BlackLivesMatter to black liberation. Haymarket Books. Taylor, M. C. (1998). How white attitudes vary with the racial composition of local populations: Numbers count. American Sociological Review, 63, 512–535. https://doi.org/10.2307/2657265 Tolnay, S. E., & Beck, E. M. (1992). Racial violence and black migration in the American south, 1910 to 1930. Amer- ican Sociological Review, 57, 103–116. Tolnay,S.E.,&Beck,E.M.(1995).Afestivalofviolence:Ananalysisofsouthernlynchings, (pp.1882–1930).University of Illinois Press. UCR. (2017). Data declaration. Uniform Crime Reports hate crime statistics, 2016. Federal Bureau of Investigation. https://doi.org/10.1037/pspp0000048 https://doi.org/10.1037/pspp0000048 https://doi.org/10.1111/1540-4560.00258 https://www.nytimes.com/2015/12/13/opinion/the-words-that-killed-medieval-jews.html https://www.nytimes.com/2015/12/13/opinion/the-words-that-killed-medieval-jews.html http://www.pewhispanic.org/2010/10/28/illegal-immigration-backlash-worries-divides-latinos/ http://www.pewhispanic.org/2010/10/28/illegal-immigration-backlash-worries-divides-latinos/ https://doi.org/10.1111/j.1745-9125.1995.tb01180.x https://doi.org/10.1177/0003122414555885 https://doi.org/10.2307/1519748 https://ssrn.com/abstract=3149103 https://doi.org/10.2139/ssrn.3149103 https://doi.org/10.2139/ssrn.3149103 http://nccs.urban.org/ http://nccs.urban.org/ https://ovc.ncjrs.gov/ncvrw2018/fact_sheet.html https://www.washingtonpost.com/news/the-fix/wp/2017/06/16/theyre-rapists-presidents-trump-campaign-launch-speech-two-years-later-annotated/?utm_term=.e279b17a93b8 https://www.washingtonpost.com/news/the-fix/wp/2017/06/16/theyre-rapists-presidents-trump-campaign-launch-speech-two-years-later-annotated/?utm_term=.e279b17a93b8 https://doi.org/10.2307/2096296 https://doi.org/10.1086/230998 https://doi.org/10.2307/2657265 746 Dugan and Chenoweth U.S. Department of Justice. (2009). Federal Bureau of Investigation. Uniform Crime Reporting data [Record-Type Files]. Inter-university Consortium for Political and Social Research [distributor], 2011-9-13. https://doi.org/10. 3886/ICPSR30764.v1 Van Dyke, N., & Soule, S. A. (2002). Structural social change and the mobilizing effect of threat: Explaining levels of patriot and militia organizing in the United States, Social Problems, 49, 497–520. https://doi.org/10.1525/sp. 2002.49.4.497 Vazquez, Y. (2010). Perpetuating the marginalization of Latinos: A collateral consequence of the incorporation of immigration law into the criminal justice system. Howard Law Journal, 54, 639–674. Vidales, G., Day, K. M., & Powe, M. (2009). Police and immigration enforcement: Impacts on Latino(a) residents’ perceptions of police. Policing: An International Journal of Police Strategies & Management, 32, 631–653. https: //doi.org/10.1108/13639510911000740 Wacquant, L. (2006). From slavery to mass incarceration: Rethinking the “race question” in the United States. In D. Macedo & P. Gounari (Eds.), The globalization of racism (pp. 94–110). Routledge. Wasow, O. (2020). Agenda seeding: How 1960s black protests moved elites, public opinion, and voting. American Political Science Review, 114, 638–659. https://doi.org/10.1017/S000305542000009X Western, B., & Pettit, B. (2010). Incarceration and social inequality. Daedalus, Summer, 8–19. https://doi.org/10. 1162/DAED_a_00019 Wright Jr., G. C. (1977). Contextual models of electoral behavioral: The southern Wallace vote. The American Polit- ical Science Review, 71, 497–508. https://doi.org/10.1017/S0003055400267415 Xie, M., & Baumer, E. P. (2019). Neighborhood immigrant concentration and violent crime reporting to the police: A multilevel analysis of data from the National Crime Victimization Survey. Criminology, 57, 237–267. https: //doi.org/10.1111/1745-9125.12204 Young, J., & Dugan, L. (2011). Veto players and terror. Journal of Peace Research, 48, 19–33. https://doi.org/10.1177/ 0022343310386269 AUTH OR BIOGRAPH IES Laura Dugan is Professor and Associate Chair in the Department of Criminology and Crim- inal Justice at the University of Maryland. Her research interests include predictors and con- sequences of extremist and hateful violence, the efficacy of violence prevention/intervention policy and practice, and the intended and unintended consequences of government actions. Erica Chenoweth is the Berthold Beitz Professor in Human Rights and International Affairs attheJohnF.KennedySchoolofGovernmentandaSusanS.andKennethL.WallachProfessor at theRadcliffeInstituteforAdvancedStudyatHarvardUniversity.Theydirect theNonviolent Action Lab at Harvard’s Carr Center for Human Rights Policy, where they are core faculty. Their research interests include mass movements, nonviolent resistance, political violence, state repression, and democracy. SUPPORTING INFORMATION Additional supporting information may be found online in the Supporting Information section at the end of the article. How to cite this article: Dugan L, Chenoweth E. Threat, emboldenment, or both? The effects of political power on violent hate crimes. Criminology. 2020;58:714–746. https://doi.org/10.1111/1745-9125.12259 https://doi.org/10.3886/ICPSR30764.v1 https://doi.org/10.3886/ICPSR30764.v1 https://doi.org/10.1525/sp.2002.49.4.497 https://doi.org/10.1525/sp.2002.49.4.497 https://doi.org/10.1108/13639510911000740 https://doi.org/10.1108/13639510911000740 https://doi.org/10.1017/S000305542000009X https://doi.org/10.1162/DAED_a_00019 https://doi.org/10.1162/DAED_a_00019 https://doi.org/10.1017/S0003055400267415 https://doi.org/10.1111/1745-9125.12204 https://doi.org/10.1111/1745-9125.12204 https://doi.org/10.1177/0022343310386269 https://doi.org/10.1177/0022343310386269 https://doi.org/10.1111/1745-9125.12259 Threat, emboldenment, or both? The effects of political power on violent hate crimes* Abstract 1 | INTRODUCTION 2 | INTERGROUP POWER CONTESTS-POWER THREAT THEORY 2.1 | Political threat hypothesis 2.2 | Emboldenment 3 | HYPOTHESES 4 | DATA AND METHOD 4.1 | Data 4.1.1 | Violent hate crime rate 4.1.2 | Federal government actions relevant to specific groups 4.1.3 | Sources for other key variables 4.2 | Methods 5 | RESULTS 5.1 | Summary statistics 5.2 | Vector autoregression models 6 | DISCUSSION ORCID REFERENCES AUTHOR BIOGRAPHIES SUPPORTING INFORMATION