Introduction

According to the 2019 Uniform Crime Report (UCR) (FBI, 2020), race and ethnicity remain the most common bias motivations in all recorded hate crime offenses in the USA, taking up nearly 58% of all hate crimes recorded. Among these cases, about 85% of offenses were racial hate crimes against non-White victims. The overwhelming proportion of racial violence against racial minorities is not a new phenomenon. The history of the USA has witnessed various forms of exclusion and suppression toward people of color — from the pre-Civil War era to the 1950s, from lynching to residential segregation (Berlet & Lyons, 2000; Levin, 2002; Turpin-Petrosino, 2009).

As one of many attempts and efforts to address ethnoviolence and empower the minorities, hate crime and related statutes were developed and mobilized by the social movements back in the 1960s to 1980s (Chakraborti & Garland, 2009; B. Levin, 2002). Sharing the collective emphasis on rights and anti-discrimination, the civil rights movement, the women’s movement, and the gay and lesbian movement were able to unite and draw attention to the violence against minorities (Jenness & Grattet, 2001). Later, the support for victims of violence was identified as a focal point in the crime victim movement (Weed, 1995), seeking criminal justice responses to victims’ civil rights. Consequently, the idea of hate crime as a social problem was prioritized within research and policy agendas.

With the stage set, both state-level and federal-level laws have been passed. In 1981, Wisconsin became the first state to pass victims’ rights legislation, followed by 42 states passing victims’ rights bills by 1989 (Weed, 1995). Since 1981, led by Washington and Oregon in 1981, states began enacting hate crime legislation. In 1993, the US Supreme Court upheld hate crime laws as constitutional. After the tragic case of Matthew Shepard, the Hate Crimes Prevention Act was signed into law by President Barack Obama in 2009, extending the biased motivation in hate crimes to actual or perceived gender, gender identity, sexual orientation, and disability.

At the state level, the protected characteristics of the victims vary a lot. While racial bias is the most common ground that all 50 states and DC have regulated already, crimes against the victim’s sexual orientation, gender, gender identity, and disability are not fully protected across states (Chakraborti & Garland, 2009; Jacobs & Potter, 2000; Lieberman & Freeman, 2009). Scholars have found that gender, disability, and sexual orientation tend to have fewer reports in nonurban areas compared to other categories that are covered earlier (B. Levin, 2009). State hate crime laws also differ from each other when it comes to which criminal acts are qualified as hate crimes, with the premise of bias motivation and intentionally targeting. Some states only cover harassment or intimidation, while other states either qualify for any offense or some particular type of offense (Jacobs & Potter, 2000).

Yet, the enactment of hate crime statutes does not mark the end of racism in the USA. The National Crime Victimization Survey (NCVS) suggested that from 2004 to 2015, residents in the USA experienced an average of 209,000 racial/ethnicity hate crime victimization each year, consisting of about 83% of total hate crime victimizations (US Bureau of Justice Statistics, 2017). In contrast, UCR only captured about 4800 racial/ethnicity hate crimes yearly, consisting of only about 2% of NCVS data.

The huge gap between these two data sources illustrates the underreporting issue of hate crimes. Research has found that victims with minority status are dramatically more likely to underreport bias-motivated victimization for various reasons, from fear to distrust of law enforcement (Culotta, 2005; Herek et al., 2000). More importantly, these personal barriers and systemic obstacles form the racialized realities for the minorities, resulting in the inability to mobilize law enforcement responses. As Feagin (2000) argued, the core racist realities are manifested in society’s major institutions, such as the criminal justice system; in turn, these institutions serve as not only a reflection but also the engine to continue resources inequalities along racial lines and ultimately maintaining and rationalizing White privilege and power in the society. In this sense, racial hate crimes — both recorded and unreported — are a form of manifestation of the racial structure in the USA. On the one hand, recorded racial hate crimes are one of the most overt expressions of racism and discrimination. On the other hand, unreported racial hate crimes generate a dark figure that not only questions the effectiveness and fairness of criminal justice system but also outlines the shadow of structural racism in the USA. Then, what marks the differences between victims who reported racial hate crimes and those who did not?

Despite the disagreement on a universal hate crime definition, two center elements construct the foundation of a hate crime in the legal statutes: a criminal act and a biased motive. It is the biased motive that distinguishes hate crimes from their parallel offenses. The responsibility of distinguishing hate crimes from other criminal actions falls on the shoulders of the individuals. As the first and crucial step to mobilize the criminal justice system (Hindelang et al., 1978), witnesses’ and victims’ notification is estimated to lead to about 80% of law enforcement personnel actions (Skogan, 1984). As such, understanding how individuals perceive and construct racial hate crimes is one of the foundational steps to solve the puzzle of underreporting racial hate crimes.

Using a factorial survey experiment with random vignette assignment, this study investigates the following two main research questions:

  • RQ1: What influences individual’s perception of racial hate crimes?

  • RQ2: How do respondents’ prejudice and beliefs influence perception of racial hate crimes?

Literature review

A crime beyond hate: a power-relation perspective

While motivation is an important part of defining a hate crime, some scholars contend that a biased motivation is not the most vital part of defining a hate crime (Gerstenfeld, 2017). Rather, the key to identifying a hate crime is the group affiliation of the victim (Chakraborti & Garland, 2009). Hate crimes are viewed as “message crime” often (Hall, 2012; Lim, 2009), which conveys a message to not only the individual victim but more so to the community that the victim belongs to (Chakraborti & Garland, 2009; Lim, 2009; Perry, 2002a). The in terrorem effect (Iganski, 2001; Perry & Olsson, 2009) — the fear and intimidation by hurting the individuals of certain groups — marks one of the arguably most important claims about hate crime (Herek, 1992; Jenness & Grattet, 2001; B. Levin, 2002; J. Levin & McDevitt, 2013), that is, hate crimes carry out more physical and psychological harm to the victims. The symbolic meaning behind a hate crime combined with its greater harm reflects the legacy of the crime victim movement, which seeks a legal response to anti-minority violence.

Such a claim is examined through a power-relation framework, which views ethnoviolence, such as racial hate crimes, as both a product of the racialized social system and a contributor to the struggle over racial positions at all levels of the racial order (Bonilla-Silva, 1997). Under such a structural view, racial hate crime is embedded in a broader pattern of oppression (Perry, 2002a; Young, 1999), which is a manifestation of the types of prejudice, marginalization, and oppression embedded within the social structure (Petrosino, 1999). They are both a product of and a tool to reinforce imbalanced power relations (Petrosino, 1999). Similarly, Young (1999) frames bias-motivated violence into five interrelated forms of oppression, with a focus on common experiences among target groups: exploitation, marginalization, powerlessness, cultural imperialism, and violence. Structural oppressions, paired with cultural foreignness, provide the context for systemic violence, which often takes on the form of hate crimes. This framework is also expanded to include a broader range of marginalized groups, such as sexual minorities (Herek, 1992), attributing hate crime to social and political factors that legitimate hate violence (Jacobs & Potter, 2000; Perry, 2002b). Perry (2002a) holds a similar view, arguing that hate crime is a mechanism of power and oppression in order to maintain and reaffirm a given social order through the creation of fear, hostility, and suspicion, which ultimately reinforces the hierarchy between the dominant group and the subordinate groups by marking the boundaries between “us” and “others.”

The “othering” process emphasizes the idea of an asymmetrical relation between victims and offenders of hate crimes. That is, the victim involves a minority status, whereas the offender is from a group of majority (Lyons, 2008). Literature has found support for a power-relation paradigm, highlighting the alignment of individual’s perception of hate crime with the legislative framework. In a mock juror study on racially motivated hate crimes, Gerstenfeld (2003) finds that White offenders are more likely to be convicted, and the certainty of guilty is higher too. Testing on the inter-racial group effect, another mock juror study reaches the same conclusion: when a White offender commits a hate crime against a Black victim, it is viewed more negatively and more severe than a Black offender White victim case (Marcus-Newhall et al., 2002). The framework remains salient when applying to other minority groups, such as sexual orientations and gender identities (Cramer et al., 2013; Mason-Bish & Duggan, 2020).

While most evidence confirms the majority offender minority victim pattern, the other side of the story has not been examined thoroughly. In one study that tests on Black-on-White incidents, Lyons (2008) observes high ratings for such incidents, suggesting a broader definition of racial hate crime that does not limit to the White-on-Black pattern. In a study with open-ended questions on hate crime, Craig and Waldo (1996) find the minority status of a victim is not associated with hate crime definitions among half of their sample.

Prejudice and bias

The other side of the “othering” process follows a social psychology approach. As one of the classic social psychology explanations of inter-group relations, social dominance theory asserts that both individual and structural determinants contribute to the creation and maintenance of group-based oppression, such as racism (Sidanius & Pratto, 1999; Sidanius et al., 2004). As a manifestation, prejudice and bias are byproducts of the social hierarchy. This approach adds insight to other structural theories by accounting for not only material and symbolic resources (Bobo & Fox, 2003; Bobo & Hutchings, 1996; LeVine & Campbell, 1972) but also individual differences in the degree of bias (Sidanius et al., 2004). As an extreme form of societal prejudice (Cramer et al., 2013), hate crimes are also affected by one’s preexisting stereotypes and prejudice against outgroup members.

Existing literature has begun to investigate the way in which preexisting prejudice and bias affect the perception of racial hate crimes. For example, research finds that respondents’ beliefs, such as bias on certain groups and attitude on punishment, influence their judgment on supporting hate crime laws (Dunbar & Molina, 2004; Steen & Cohen, 2004). Respondents with higher racial prejudice are less supportive to hate crime laws than their counterparts (Dunbar & Molina, 2004). Blame attribution is also related to bias. Individuals with higher racial prejudice tend to find hate crime victims more blameworthy (Cramer et al., 2010; Marcus-Newhall et al., 2002; Rayburn et al., 2003). However, other literature has not found a similar effect. For instance, Gerstenfeld (2003) does not find a significant difference among participants with high and low levels of racism in performing jury duty and assigning sentences. Similarly, negative attitudes towards sexual minorities do not influence individuals’ rating for the certainty and seriousness of a hate crime (Lyons, 2008).

Gaps in the literature

While previous studies have revealed several important components when constructing hate crimes, several issues need further examination. First, there are few examinations on individual’s perception of racial hate crimes. Previous studies focus heavily either on evaluating the seriousness of hate crime (Cramer et al., 2010; Gerstenfeld, 2003; Marcus-Newhall et al., 2002; Rayburn et al., 2003) or testing on the supportiveness of hate crime legislation (Dunbar & Molina, 2004). These findings are based on a general assumption — there is no variation among individuals regarding what is a hate crime. Yet, such an assumption has to be examined very carefully. In one of the few studies that examined such a relationship, Lyons (2008) has revealed a strong effect of the presence of hate messages on recognizing hate crime. The presence of hate messages, such as racial slurs, offers evidence of the bias motivation, which concretizes the elusive component in hate crime legislation. As the scenario gets more complicated, respondents’ own characteristics come into play. A qualitative examination of hate crime definitions has found great variations among respondents with different demographic characteristics (Craig & Waldo, 1996).

Second, the lack of empirical evidence on potential differences of inter-racial and intra-racial hate crime patterns hinders the understanding of how hate crime legislations are perceived. Among few studies that examined racial power relations, the results are not conclusive. Marcus-Newhall et al. (2002) have found greater certainty of guilty and sentencing for cases with White perpetrators and Black victims than Black-White or Black-Black scenarios. Contrarily, another study has not observed significant differences between White-Black and Black-White patterns (Lyons, 2008). Prior literature often focuses on a binary comparison, i.e., White and Black, neglecting subtle differences among minorities as victims and offenders.

Third, although prior studies have explored some aspects of prejudice and racial hate crimes, the effect of preexisting bias on categorizing a criminal act as hate-motivated remains unclear. Previous studies have a heavy focus on the linkage of racial bias and support of hate crime laws (Dunbar & Molina, 2004; Steen & Cohen, 2004). Yet, how individuals view the legitimacy of hate crime laws does not necessarily ensure the effectiveness of the legal statutes (Dunbar & Molina, 2004). Blame attribution research, on the other hand, also fails to establish such a pathway (Cramer et al., 2010). Neither of these two major themes in the field is directly related to the certainty of racial hate crimes.

Moreover, college students are the most common samples in previous studies (e.g., Lyons, 2008; Marcus-Newhall et al., 2002; Rayburn et al., 2003; Steen & Cohen, 2004), which limits the generalizability of the findings. Admittedly, experimental designs with convenient samples may not necessarily compromise external validity. The highly homogeneous samples, such as college students, are likely to constrain further analyses in regard to diverse demographic status.

Current study

This study aims to bridge the power-relation framework with prejudice and bias studies. Specifically, two research questions are examined:

  • RQ1: What influences individuals’ perception of a racial hate crime?

  • RQ2: How do respondents’ prejudice and beliefs influence perception of a racial hate crime?

Incorporating significant determinants from existing literature, the first set of hypotheses examines incident characteristics, with a focus on victims and offenders’ race (Gerstenfeld, 2003; Lyons, 2008; Marcus-Newhall et al., 2002; Steen & Cohen, 2004). Other incident characteristics, including the presence of hate messages (Lyons, 2006, 2008), violence type, and violence level, are included as controls.

  • Hypothesis 1: The races of the victim and offender influence the perception of a racial hate crime.

Adopting a power-relation framework, this paper emphasizes the power dynamic between victims and offenders in the incidents while including both inter-racial and intra-racial patterns. The examinations beyond a simple Majority-on-Minority pattern supplement prior research in the field. Moreover, the current study also extends investigations to race/ethnic groups beyond White and Black.

  • Hypothesis 1a: Respondents are more likely to recognize a racial hate crime when the incident involves a racial majority offender and a racial minority victim.

  • Hypothesis 1b: Respondents are less likely to recognize a racial hate crime when the incident involves a racial minority offender and a racial majority victim.

  • Hypothesis 1c: Respondents are less likely to recognize a racial hate crime when the victim and offender both are racial minorities.

  • Hypothesis 1d: Respondents are less likely to recognize a racial hate crime when the victim and offender belong to the same racial group.

To integrate a social psychology perspective, the following hypothesis is tested:

  • Hypothesis 2: Respondents' prejudice and beliefs influence the perception of a racial hate crime.

Specifically, I include two scales in the instrument. The first one is Right-wing authoritarianism (RWA hereafter). RWA was first developed by Altemeyer (1981), capturing conventionalism, authoritarian aggression, and authoritarian submission. RWA is a strong predictor of prejudice, ethnocentrism, and homophobia and is broadly used in social studies (Altemeyer, 1998; Duckitt & Sibley, 2010; Sidanius & Pratto, 1999). Research has found that people with high authoritarianism were prejudiced against racial minorities (Akrami et al., 2000), women (Sibley et al., 2007; Whitley, 2001; Whitley & Aegisdottir, 2000), lesbians and gay men (Laythe et al., 2001; Whitley, 1999, 2001), people with disabilities (Crowson et al., 2013), and immigrants (Akrami et al., 2000; Davidov et al., 2008). As such, RWA is capable of detecting a broad range of prejudice.

  • Hypothesis 2a: Respondents with higher Right-Wing Authoritarianism (RWA) agreement are less likely to perceive a racial hate crime.

The second scale I adopted is the Perception of Police Scale (POPS hereafter), which is designed by Nadal and Davidoff (2015) to measure general attitudes toward police and perceptions of police bias. It has long been recognized that attitude toward law enforcement influences cooperation with the criminal justice system, especially among racial minorities (Brown & Reed, 2002; Huebner et al., 2004; Nadal et al., 2017; Peck, 2015; Slocum, 2018). A more positive attitude toward police is proved to facilitate better cooperation with the criminal justice system and compliance with the law (Bradford & Myhill, 2015; Tankebe, 2013; T. Tyler & Fagan, 2008; T. R. Tyler, 2004; T. R. Tyler & Huo, 2002). As the first step to initiate criminal justice responses, defining racial hate crimes may be considered under the broader notion of compliance with the law, which is sensitive to the influence of the attitude toward police.

Moreover, recent highly publicized police excessive force against people of color — for instance, Breonna Taylor and Michael Brown — and may also have a negative impact on individual’s attitude toward police, especially among the marginalized populations. The negative experiences related to police interaction could be unevenly distributed among racial/ethnic groups and lead to lower compliance with the law and criminal justice system in general.

  • Hypothesis 2b: Respondents with a more positive attitude in the Perception of Police Scale (POPS) are more likely to perceive a racial hate crime.

Data and methods

Factorial survey experiment

This study employs a factorial survey experiment with randomized vignette assignments to obtain insights into respondents’ judgment principles. Factorial survey experimental design is a multidimensional method that combines survey research and experimental research (Auspurg & Hinz, 2014; Lyons, 2008; Rossi & Anderson, 1982; Rossi & Nock, 1982). In most cases, a factorial survey uses descriptions of fictional situations (vignettes) as stimuli, followed by survey questions that ask respondents to evaluate scenarios. Within each vignette, a number of important characteristics (dimensions) systematically vary by type or degree (level).

As a combination of survey methods and experimental research, factorial survey experiments have a number of advantages. First, by introducing randomly assigned vignettes, factorial survey experiments guarantee internal validity (Auspurg & Hinz, 2014; Rossi & Nock, 1982). That is, the respondents’ reactions only reflect variations in the vignettes only, reducing the correlation between respondent characteristics and the hypothetical scenario (Auspurg & Hinz, 2014). Second, the form of a survey study can be applied to relatively heterogeneous populations fairly easily, which increases the generalizability and external validity of the study. Traditional experiments often suffer from homogeneous groups of participants, such as college students or self-selected samples. The difficulties in recruiting relatively diverse participants also limit the number of factors that an experiment could test. A survey study, on the other hand, is able to avoid the shortcoming by attracting participants from large and random population samples, which makes experiments and survey methods compatible (Auspurg & Hinz, 2014). Moreover, the factorial approach opens opportunities to investigate more realistic complexity than traditional survey methods because of the introduction of vignettes that vary in dimensions and levels. Due to the common underreporting issue of hate crime, observational data may not be able to exhaust all possible situations in hate incidents. The factorial survey, on the contrary, can expand reality by presenting vignettes that are not recorded in reality. It does not require pre-exposure to certain types of victimization to estimate the perception or willingness to report.

Vignette design

This study focuses on violent crimes that involve crimes against persons. Hypothetical scenarios are presented and randomly assigned to participants, followed by questions to evaluate described incidents. The dimensions include the victim’s race, offender’s race, presence of hate message, and type and level of violence (Table 1). Other dimensions of the incident, such as age, gender, and location, are omitted intentionally in order to reduce noise. For example, both victim and offender are given gender-neutral names so as to avoid gender as an extra dimension. Similarly, victims and offenders are presented as strangers in the vignette.

Table 1 Vignette design and distributions (n = 2272)
Full size table

In order to minimize the dimensions in the vignettes and provide more reasonable information, race and ethnicity are combined into one single dimension with three levels (White, Black, or Latino). The limited number of levels reflects the most commonly targeted racial and ethnic groups (FBI, 2020). The presence of a hate message refers to whether a racial slur is outspoken by the offender during the incident. Types of violence include verbal threats and physical assault. Levels of violence consist of no injury, slightly injured, and severely injured. The level of violence in a vignette depends on the type of violence. For instance, the verbal threat is only associated with no injury in the vignette, whereas physical assault can result in all three levels of violence. Further, to exhaust all possible combinations of the incident, within-group incidents (e.g., white offender and white victim) are also included in the vignette universe.

The total number of possible vignette combinations is 72. Each respondent receives a randomly assigned vignette, followed by questions related to the vignette. All the vignettes are evenly distributed.

An example vignette is presented below. In this vignette, the offender is African American, and the victim is Hispanic/Latino. It also includes the presence of hate messages (i.e., yelling racial slurs) and type and level of violence (i.e., physical assault and severely injured).

Dakota is African American. Sam is Hispanic.

One day, Sam was walking in the street. As Sam passed near Dakota, Dakota beat Sam up while yelling derogatory racial slurs. Sam was severely injured and stayed in the hospital for a whole week.

Other measures

Perceived racial hate crime

The outcome of this study is whether an incident is perceived as a racial hate crime. Following the legal definition, I ask two questions to capture this measure. One question measures the degree of agreement on the vignette being a crime, and the other question evaluates the degree of certainty of racial motivation. Participants answer both questions based on the vignette on a 5-point Likert scale. Then, I combine and dichotomize the answers. When the participant agrees or strongly agrees to both questions, the variable is marked as a racial hate crime. Otherwise, it is marked as not a racial hate crime. This seemingly complicated measure is designed to increase the accuracy of the measurement. It is reasonable to assume that individuals are unlikely to explicitly call an incident a racial hate crime unless they are fairly certain about both the incident being a crime and motivated by racial bias. Not only does the structure of the questions mirrors the general commonality of all hate crime legislation, but it also avoids the usage of the term “hate crime,” which could trigger certain reactions among some respondents.

Right-wing authoritarianism

Participants answer a short version of RWA with 15 statements, which have been approved to function similarly to the original scale but with fewer items and less extreme wordings (Zakrisson, 2005). Participants are asked to rate their agreement to each statement on a 5-point Likert scale. The mean of the 15 items is calculated and adopted as the score for this measure. A higher score for this measure indicates higher agreement of RWA, with a maximum score at 5. The reliability coefficient of the RWA is 0.86.

Perception of Police Scale

This measure consists of 12 items assessing the general attitudes toward police and perceptions of police bias (Nadal & Davidoff, 2015). An average score is calculated and adopted in this study. A higher score for this measure indicates a more positive attitude toward police, with a maximum score of 5. The reliability coefficient of the POPS is 0.94.

It is worth mentioning that the correlation between RWA and POPS is moderate at 0.42. Factor analysis is conducted for all items included in the instrument. Results show that each of these two scales includes two factors without overlapping (factor loading > 0.5), suggesting the dimensions captured in RWA and POPS are different. Further, models in this project all passed multicollinearity tests (Mean VIF = 1.09, maximum VIF = 1.24). As such, both RWA and POPS are included in the analyses.

Demographics

Demographic information includes respondent gender, sexual orientation, race/ethnicity, religious belief, age, education level, self-reported social-economic status, and previous violent crime victimization.

Data collection

I utilize an online crowdsourcing service Mechanical Turk (MTurk hereafter) to recruit participants and collect data. MTurk is an online marketplace where individuals or businesses can outsource jobs or tasks to a large pool of people. Potential workers can pick and choose any task to complete for the compensation the employer offers (Mason & Suri, 2012). It is a popular method of recruiting paid participants for surveys and psychological experiments due to its cheap availability and relatively diverse population (Buhrmester et al., 2011).

There are several benefits of using Mturk instead of traditional recruitment strategies. First, MTurk has a large worker pool. Research has shown that MTurk became one of the most widely used online subject pools with about 500,000 adults from 190 countries (Behrend et al., 2011; Paolacci & Chandler, 2014). Second, the cost of using MTurk is relatively low, considering the speed at which one can collect data. In one study, researchers obtained 250 completed surveys within 24 h at the cost of $350 (Renzetti & Lynch, 2018). Finally, MTurk offers a better sample than other convenient and student samples that are often used in factorial survey experiments in related. It has long been recognized that college student samples are not representative of the general population, which compromises the generalizability of the results. MTurk workers are found to have a median age of 30, with the overrepresentation of college graduates, underemployed, middle to lower-middle-class, and white population (Paolacci & Chandler, 2014). Compared to a nationally representative sample, MTurk generates similar results (Mullinix et al., 2015).

During the participant recruitment and data collection process, I post a short description of the study and offered $1.00 compensation for survey completion. Eligible participants must be US residents and 18 years old and above. As MTurk requires for minimum working age of 18 years old, only one additional sample filter for US residency is requested to ensure the eligibility of the participants. Participants are redirected to an anonymous Qualtrics link to complete the consent form and take the online survey. The data collection of 2635 responses concludes on May 16, 2020, within 31 hours, with an average survey time of around 26 minutes.

Participants

Among the 2635 responses collected through MTurk, I identify 107 (4.06%) invalid responses through multiple measurements, including direct questions and archival items. Direct questions include age, US residency, self-report of carefulness, and an instructed item. Although US residency is set as a filter that qualifies MTurk workers for this study, it is used as a screening question at the beginning of the survey. Age is used both as a control variable and as a screening item to exclude those either below 18 years old or give extreme numbers, such as “999.” A question on self-report of carefulness asks the respondent whether they have paid attention during the survey process. If not, the answer is excluded from the final sample. Archival items include three measures. First, the length of time a participant takes to complete the survey signals a potential inattentive response. While the threshold for a minimum length of time is not universally established, I adopt Huang et al.’s (2012) approach, which classifies the response being insufficient with less than 2 s for personality questions or attitudinal items. There are 129 items to be rated or answered in the survey. If the respondent completes the survey within less than 250 s, it is classified as an invalid response. Second, responses with the progress of less than 60% are removed from the dataset (Renzetti & Lynch, 2018). In addition, Qualtrics (n.d.) marks a spam response when multiple identical responses are submitted from the same IP address within a 12-h window. Responses with the Spam flag are removed from the final sample. After listwise deletion of the identified invalid responses and missing, the final dataset consists of 2272 participants.

Table 2 presents descriptive statistics of respondent demographics and attitude scales. The demographics of the sample roughly mirrored that of the US population: 65.98% Non-Hispanic White, 16.68% Black, 6.78% Hispanic/Latino, 8.01% Asian, and 2.55% other; 54.09% male, 45.33% female, and 0.57% other gender identities; 79.36% heterosexual, 3.87% homosexual, 15.98% bisexual, and 0.79% other sexual orientations. About 18.35% of respondents identified as not religious. The average age of the sample is about 38.54 years old, ranging from 18 years old to 79 years old. The sample also has an average of 5.62 out of 9 self-identified SES. Additionally, the sample is highly educated, with only 0.36% of respondents reporting less than high school educational obtainment. The demographic composition is also in line with the previous findings (Paolacci & Chandler, 2014): median age around 34, with the overrepresentation of high school graduates and middle class.

Table 2 Descriptive statistics on respondent demographics and attitude scales (n = 2272)
Full size table

To ensure that the randomization procedure was effective in assigning balanced groups, I conduct balance tests on respondents’ demographic variables using Chi-square tests (race, gender, sexual orientation, religious or not, and any previous violent victimization experience) and one-way ANOVA (education level, age, and self-identified SES) (Table 3). Among all demographic variables tested, the dimensions of the victim’s race differ significantly with regard to educational levels (F = 2.31, d.f. = 7, p < 0.05), showing a neglectable effect size at 0.007. This observation is probably due to the overrepresentation of high school graduates and uneven distribution of education levels in the sample compared to the US population. An additional bivariate logistic regression model is estimated as a sensitivity check, revealing that the inclusion of education level does not alter the substantive results and conclusions presented below.

Table 3 Balance check by demographics for vignette conditions (n = 2272)
Full size table

Analytic strategy

Incident model

A binary logistic regression model is tested for the incident model. Model 1 includes the incident variables provided in the vignette. This model is used to test whether there is any significant association between incident characteristics and the outcome.

Attitude model

Similarly, to parse the effect of respondent’s prejudice and beliefs, a binary logistic regression model is estimated. RWA and POPS are introduced to the incident model. Model 2 includes the incident variables, RWA, and POPS.

Moderation analyses

To investigate whether the perception of racial hate crimes follows a power-relation framework, the moderation term of offender’s race and victim’s race is introduced to Model 1 and Model 2. Besides reporting the odds ratios, pairwise comparisons are performed to detect significant differences across patterns. Additional margins tests are performed for better visualization.

Results

Table 4 presents four binary logistic regression models, predicting the likelihood of perceiving an incident as a racial hate crime based on incident characteristics and two attitudinal scales.

Table 4 Binary Logistic Regression models of racially motivated hate crime indicator on incident variables and attitude scales
Full size table

Incident model

Model 1 demonstrates that incident characteristics, including offender’s race, victim’s race, presence of hate message, and violence level and consequence all significantly influence the likelihood of the outcome. Whether a racial slur is explicitly expressed by the offender has the strongest positive effect among all predictors, significantly increasing the odds ratio of perceiving a racial hate crime by 2.541.

Yet, the effects of other variables are only significant in comparisons between certain groups. For violence level and consequences, only physical harm with slight injury significantly increases the likelihood of perceiving a racial hate crime compared to a verbal threat without injury. None of the other pairs makes a difference.

In regard to offender and victim’s race, incidents with a Black or Latino offender reduce the likelihood of perceiving a racial hate crime by odds ratios of 0.680 and 0.798, respectively, compared to cases with a White offender. Similarly, a Black victim increases the likelihood of perceiving a racial hate crime by 1.272 times when all other incident characteristics are held constant. Such an effect is not observed when comparing cases with a Latino victim to a White victim. The inconsistency of the two racial minority groups is also found in the comparison between cases with a Latino victim and a Black victim. That is, a Latino victim reduces the likelihood of perceiving a racial hate crime by a factor of 0.801 compared to cases with a Black victim.

Attitude model

Along with incident variables, Model 2 includes two attitudinal scales. The results suggest that a one-unit increase in POPS score significantly increases the odds ratio of recognizing a racial hate crime by 1.473, all others constant. RWA score, however, does not significantly influence the outcome.

While Model 2 highly resembles Model 1 in terms of the results on incident variables, whether the victim is Latino or Black does not make a significant difference. Despite the fact that the directions of the effect do not change, Latino victims do not significantly decrease the probability of the respondent perceiving a racial hate crime.

Moderation analyses

Model 3 and Model 4 present two binary logistic regression models with a moderation term — offender’s race by victim’s race. Model 3 includes incident variables and the interaction term. Model 4 includes two attitudinal scales on the basis of Model 3. The general results for both models are highly similar: all incident variables and POPS are statistically significant in the same direction. As Model 4 has a better model fit, the moderation effect is discussed based on the results in Model 4.

Figure 1 illustrates the predicted probability of perceiving an incident as a racial hate crime jointly considering offender and victim’s race, holding POPS and RWA at their means. For the purpose of illustration, incidents characteristics are further held constant with the presence of hate message and a verbal threat without injury. Despite the fluctuations in the predicted margins as the other incident characteristics change, the following observations hold.

Fig. 1
figure 1

Predicted Margins on Racial Hate Crime Perception: Offender and Victim Races (Model 4). Note: All predicted margins are significant at .001

Full size image

Intra-racial incidents

The graph shows that when offender and victim are from the same racial group, the predicted probabilities of the outcome are at their lowest across all three racial groups being tested. Results suggest that White-on-White cases are not statistically different from Minority-on-Minority intra-racial incidents (Table 5).

Table 5 Pairwise comparison of predicted margins on racial hate crime perception: offender and victim races (Model 4)
Full size table

Interestingly, when comparing Minority-on-Minority intra-racial cases, Black-on-Black incidents receive a lower predicted probability at 0.282 to be recognized as a racial hate crime while Latino-on-Latino incidents have a probability at 0.380, which equals to a 34.8% difference. The predicted margin of the Black-on-Black pattern is the lowest across the whole project.

Inter-racial incidents

In the inter-group incidents, the Majority-on-Minority pattern is tested by White-on-Black and White-on-Latino. First, using White-on-White incidents as the baseline at 0.355, a racial minority victim strongly increases the probability of perceiving a racial hate crime. Specifically, White-on-Black incidents have the highest probability of being defined as a racial hate crime across all pairs at 0.732, whereas White-on-Latino cases at 0.596. However, significant differences are observed between White-on-Black and White-on-Latino cases. When holding RWA and POPS at their means, White-on-Black incidents with the presence of a hate message and verbal threat without injury are about 1.225 times more likely to be recognized as racial hate crimes compared to White-on-Latino cases under the same condition.

Second, when the victim is a racial minority, a White offender significantly increases the likelihood of identifying a racial hate crime. Comparing to Minority-on-Minority inter-racial cases, the cases with a Majority-on-Minority pattern have a significantly higher chance of being identified as a racial hate crime. For instance, White-on-Black cases have the highest predicted margins to be recognized as a racial hate crime at 0.732, whereas a Latino-on-Black case has a significantly lower probability at 0.621. Similarly, White-on-Latino cases are 1.160 times more likely to be categorized as racial hate crimes compared to Black-on-Latino incidents.

Third, after examining Minority-on-White patterns, no significant difference is found between Black-on-White and Latino-on-White incidents. However, when comparing Minority-on-White patterns with White-on-Minority patterns, Black-on-White cases have a significantly lower probability of being recognized as racial hate crimes compared to White-on-Black cases by over 23%, whereas only neglectable difference is found between Latino-on-White and White-on-Latino cases. Black-on-White incidents also rank the highest among cases with a Black offender at 0.595, demonstrating consistency with findings regarding White-on-Minority patterns.

Moreover, inter-racial Minority-on-Minority cases — namely Black-on-Latino and Latino-on-Black — are statistically different in terms of predicted margins for the outcome. Latino-on-Black cases are 1.208 times more likely than Black-on-Latino cases to be perceived as a racial hate crime. Notably, this ratio is very close to that of the White-Black dyads, with a merely 2% difference.

Other incident variables

Similar to findings in Model 1 and Model 2, the presence of hate message remains the strongest predictor of a higher probability of racial hate crime perception. Compared to incidents without a verbal cue, the presence of hate message is associated with 2.813 times more likely to perceive a racial hate crime, holding all other covariates constant. Incidents with physical harm and slight injury increase the probability of assigning a racial hate crime by a factor of 1.403 compared to a verbal threat without injury, which is also the only significant pair of levels under this variable.

Discussion

Using data from a factorial survey experiment, this study examines the determinants that influence the perception of racial hate crimes among US adults recruited from an online crowdsourcing service. The results offer support to an integrated model that both the power dynamics between the victims and the offenders and the preexisting prejudice and beliefs of the respondents play significant roles in the process of constructing a racial hate crime. For the most part, respondents react to incident characteristics in ways suggested by the power-relation framework even after considering one’s attitude toward police, whilst results reveal some important variations by the racial groups that victims and offenders belong to.

First of all, interactions between the offender’s race and victim’s race indicate increased sensitivity to inter-racial incidents. For all three racial groups tested, inter-racial incidents all have a higher probability of being perceived as racial hate crimes compared to intra-racial incidents. However, rather than sticking to a Majority-on-Minority pattern, respondents seem to view incidents with a Minority-on-Majority pattern as equally problematic. For instance, there is no significant difference observed between a Latino-on-White incident and a White-on-Latino incident. Despite the observed variance in incidents between White and Black, the predicted margins for Black-on-White incidents are the highest among the cases with a White victim. Minority-on-Majority cases also indicate a higher probability of being assigned as a racial hate crime than Minority-on-Minority inter-racial incidents. According to social dominance theory (Pratto & Stewart, 2011; Sidanius & Pratto, 1999), the Minority-on-White incidents may have broken the racial hierarchy in society, which creates an increased sense of threat from the minority group. As a result, racial minority offenders are equally likely to be called out for racial hate crimes. These findings are consistent with prior literature that tests on White-Black dyads (Gerstenfeld, 2003; Lyons, 2008; Zaykowski, 2010), suggesting that rather than supporting a Majority-on-Minority pattern of racial hate crime, respondents are more sensitive to incidents with offenders and victims from different racial groups, regardless of the direction in power relations in the incidents.

While inter-racial incidents have a higher likelihood to be designated as racial hate incidents in general, examining the racial identities of the offender and victim reveals more nuances. Among all inter-racial incidents, respondents react most strongly to incidents with a White offender. This observation parallels the spirit of hate crime legislation and antiviolence social movements that advocate for addressing ethnoviolence and seeking racial justice. It also supports the power-relation perspective that views racial hate crime as one form of racial oppression that is most commonly seen in White-on-Minority cases (Perry, 2002a; Young, 1999).

Yet, the power dynamics between racial groups are also found between racial minorities. On the one hand, the findings demonstrate the existence of the stereotype of Blacks being the victim of racial hate crimes. In the cases with a White offender, the predicted probability of defining a racial hate crime is largely dependent on who the victim is. The results indicate that White-on-Black cases are more than 23% more likely to be perceived as a racial hate crime than White-on-Latino cases when all other conditions are the same.

The idea of a Black victim being more stereotypical may be traced back to Black history in the USA, such as slavery, lynching, residential segregation, and mass incarceration (Berlet & Lyons, 2000; B. Levin, 2002; Turpin-Petrosino, 2009). Recent social movements around Black Lives Matter (BLM) may also play a role in bringing public attention to racial inequalities in the USA, especially discriminatory actions against Blacks. Although the data collection of this project was concluded on May 16, 2020, just before the Murder of George Floyd and protests afterward, the exposure to recent Anti-Black incidents with excessive violence, such as the killing of Breonna Tylor in March 2020, along with the years-long efforts surround BLM, have made it impossible to overlook the oppression and racism toward Blacks. As some research has shown, one of many social movement outcomes is public resource mobilization (Mcveigh et al., 2003; Snow et al., 2007). Perhaps hate crime reporting — at least increased perception of racial hate crimes — also represents a successful outcome of social movements around racial justice.

The image of Blacks being stereotypical victims in racial hate crimes may also overshadow the recognition of other racial minorities as racial hate crime victims. For instance, White-on-Black incidents exceed White-on-Latino cases by 23% in predicting the likelihood of assigning a racial hate crime, suggesting that Latino is viewed as less of a minority group than Black. Such observation is supported by the 21% higher probability of recognizing Latino-on-Black cases in comparison to Black-on-Latino ones.

There is more to the complicity of Latino. That is, the racial position of Latino seems to be relative. On the one hand, respondents recognize the minority status of Latinos — or “less majority” status — when using Whites as a reference. In addition to the aforementioned White-on-Minority examples, Latino offenders also reduce the probability of racial hate crimes in anti-Black incidents by 18%. On the other hand, the difficulty in distinguishing Latino from White in some results even questions whether Latino is viewed as a minority group at all. The results reveal the invariance in the outcome when comparing White-Latino incidents, which is even insignificant compared to Minority-on-Minority inter-racial incidents. The double standard in perceiving Latino as a minority group highlights the idea of them being invisible, which perpetuates the disadvantages of Latino in mobilizing public resources and seeking criminal justice responses.

One possible explanation for the ambiguity in Latinos’ minority status may come from a different dimension of their disadvantages, that is, the perceived cultural foreignness (Zou & Cheryan, 2017). Under the Racial Position Model (Zou & Cheryan, 2017), Latinos may be perceived as deviating from American mainstream culture for being stereotyped as foreigners on top of the relatively inferior racial position in the USA, whereas Blacks mainly experience the latter dimension of prejudice. As such, the usage of a racial slur as the hate message in the vignettes may have only triggered the superior-inferior dimension that signals more strongly toward Blacks. Future research will benefit from a more comprehensive design that includes cultural foreignness as a separate stimulus.

For the two attitudinal scales, only POPS has a significant effect in perceiving racial hate crimes. The different effects of these two attitudinal scales have several implications. First, the positive association between an individual’s attitude toward police and racial hate crime perception indicates that cooperation with law enforcement and compliance with the law may begin earlier than making the decision of reporting to the police. In other words, certainty about crime and agreement with existing legal statutes seem to be under the influence of attitude toward police as well. While this observation is somewhat related to legal cynicism (Adriaenssen et al., 2019; Sampson & Bartusch, 1998), this effect has not been captured in prior literature with regard to racial hate crimes to my knowledge. Future research would benefit from exploring more directly how attitudes toward police translate into perceiving and reporting racial hate crimes and crimes in general.

Second, the insignificance of RWA may be a result of the measurement itself. As Altemeyer (1981, 1998) points out, people high in RWA tend to see outgroup members as disrupting the social order that they believe in. They also have a tendency to protect and maintain traditional values (Altemeyer, 1981, 1998; Whitley, 1999; Zakrisson, 2005). However, this tendency does not necessarily result in physical violence or support for violence. On the opposite, expressing prejudice is often used as a means to stop physical violence (Whitley, 1999). To maintain the social order to their own beliefs, people with high RWA scores may take the vignette incident as a wrongful act itself. As such, the effect of RWA is not prominent in the models. Despite the insignificance, findings show that more agreement with RWA results in a lower probability of recognizing a racial hate crime. This general direction fits the hypothesis, although the effect is not strong in the models. Besides, despite the acceptable results from post-estimation of the models and factor analysis, it is still possible that the variance of RWA has been explained by POPS because one’s attitude toward police could be a result of authoritarian submission to some extent.

Among all incident variables, the explicit expression of racial bias remains the strongest predictor of the outcome. This finding is in line with previous research: verbalized racial bias is one of the essential components that distinguish a hate crime from its parallel offense, mirroring the spirit of legislation (Gerstenfeld, 2003; Lyons, 2008; Rayburn et al., 2003, 2003). As one of the most common indications of bias motivation (see (Department of Justice, 2020a, 2020b, 2021), the presence of hate messages also offers evidence of the bias motivation, which concretizes the elusive definition of hate crimes (Cramer et al., 2013).

Violence level and consequences, on the other hand, do not seem to influence the outcome evenly. Only incidents with physical harm with slight injury are perceived differently from the verbal threat without injury. Further examinations that separate violence level and consequences also support this conclusion. The unexpected findings may be a result of how this project measures the outcome. Most of the previous studies either examine the seriousness of an incident or the willingness to report (Lyons, 2007; Skogan, 1977, 1984; Zaykowski, 2010). The effect of violence level and consequences may have a stronger effect in determining reporting behaviors. Further, the presence of a hate message may also function as verbal violence, which eventually directs the respondent’s attention away from the elements of physical harm and injury.

In conclusion, this study finds empirical evidence to establish a link between the status of incidents, respondents’ beliefs and prejudice, and the perception of racial hate crimes. More importantly, this study integrates a power-relation perspective with a social psychological approach, proposing an inter-racial hate crime pattern with racial disparities. It is worth noting that this paper does not claim to offer a general theory of racial hate crimes; rather, it offers one way to critically examine how racial hate crimes are constructed through the imbalance of power dynamics. Future research would benefit from expanding the examination to other minority groups, such as sexual orientation and religious affiliation, which would offer a deeper understanding of the power-relation perspective as well as to cover the most common biased motive for hate crimes in the USA. A subgroup analysis would also advance this study by testing how respondent’s own racial identity influences their perception of racial hate crimes. To the extent that individual’s definitions of racial hate crime are important vehicles to mobilize criminal justice responses, they are also central to better policies that address racial bias both in legislation and practices.