Hate crime offending among European youth: Prevalence and risk factors

Strain and deprivation

The theoretical framework of relative deprivation and frustration-aggression supported by Hovland and Sear (1940) was one of the first explanations for hate crime (Díaz-Faes and Pereda, 2022: 944). Merton's (1938) concept of anomie has become a leading frame in the analysis of various forms of hate crime (Walters, 2011) and racist extremism (Blazak, 2024). Strain theory sees the commission of hate crimes as a signal to outside groups that are perceived as a threat to the achievement of culturally accepted goals (Hall, 2015; Sheppard et al., 2023).

Historically, perceptions of relative deprivation or a sense of impending economic hardship have been viewed as potent explanatory factors for hate crimes. However, the evidence on the poverty–hate crime link is limited and partially contradictory (Craig, 2002; Hall, 2015; Walters, 2011). Several studies suggest that there is no association between economic factors and hate crime offending or bigotry (Green et al., 1999a; Green et al., 1999b). Van Kesteren (2016) found no independent association between socioeconomic factors and experiences of hate crime victimization in Europe. Research on right-wing extremism in Germany has found strong correlations between unemployment and extremism (Falk et al., 2011). A more recent study on anti-gay hate crime in New York City (Mills, 2021) found that more disadvantaged communities tend to report more hate crime, supporting strain theory. In a qualitative study, the interviewed extremists were not characterized by absolute poverty or relative deprivation (Pisoiu, 2014), but this study focused on individuals who had gained relatively prominent positions among right-wing and jihadist movements. Over a decade ago, Chakraborti and Garland (2012) concluded that we still lack sufficient research on how economic marginalization impacts the likelihood of committing hate crimes and that ambiguity persists today.

On the other hand, general strain theory (GST; Agnew, 1992), with its much broader definition of strain, has shown its usefulness in explaining hate crimes among youth in Finland (Näsi et al., 2016). Recent analysis of an internet survey on propensity to support far-right ideology found support for the role of economic hardship and negative life events, as suggested by GST (Skoczylis and Andrews, 2022). It should be noted that Agnew (2010) himself has made a strong case that GST may be applied to terrorism.

Social learning, moral attitudes, and emotions

According to social learning theory (Akers, 1973), socially acquired values and attitudes explain involvement in criminal behavior. As applied to hate crime, this theory suggests that socially learned prejudices explain variations in the risk of committing hate crime (Craig, 2002; Díaz-Faes and Pereda, 2022; Lantz et al., 2024). A recent article on the growth of right-wing extremism in Canada through the lens of structured action theory shows how broader social, cultural, and political patterns create ‘permission to hate’ (Perry and Scrivens, 2018), highlighting the crucial importance of the context in which individuals learn acquiring particular definitions of the situation supportive of hate crime. Mills et al. (2021) found an integrated social control–social learning model useful in explaining radicalization and extremist violence. In the study of hate crime and extremism, an emerging problem is the ubiquitous nature of learning sources; the internet and social media are ever-present and provide ongoing access to like-minded peers (see, e.g., Kaakinen et al., 2020; Williams, 2021).

The role of moral emotions, particularly shame, remains unexplored in hate crime offending. In addition to learned values and attitudes, criminal behavior is influenced by moral emotions, such as shame (Svensson et al., 2013; Tangney et al., 2007). Although shame originates from breaching shared moral norms, individuals vary in their experiences across situations. Those prone to shame are more deterred by it (Wikström, 2014). Diaz-Feas and Pereda (2022, 946) concluded their review of bias-motivated violent offending by stating that (Wikstrom's) situational action theory and its moral background offer a good explanation for the perpetration of discriminatory violence in specific cultural contexts, ‘given the key role of stereotypes, prejudice, and attitudes’.

Social control, self-control, and specialization

Walters (2011) argued that economic strain theories cannot adequately explain hate crime offending and suggested that Gottfredson and Hirschi's (1990) general theory of crime (also known as self-control theory) should be used in the study of individual-level predictors of hate crime propensity. Since, according to the general theory of crime, most crimes are the result of poor self-control, hate crimes should accordingly be committed mostly by such persons. Walters (2011) found that self-control was able to explain thrill-seeking and defensive hate crimes but not retaliatory or mission-driven types (also Sheppard et al., 2023). In a nationally representative study of 15–16-year-old Finnish students, self-control was found to be an important factor correlated with self-reported hate crime offending, together with elements of GST and social control theory (Näsi et al., 2016; see also Sheppard et al., 2023).

Consistent with the general theory of crime, various studies have shown that hate crime offenders do not specialize but engage in a wide range of criminal behaviors (Messner et al., 2004; Roxell, 2011; Walters, 2011). Messner et al. (2004), for example, tested two theoretical models—specialization and versatile offender—to determine whether prejudice (specialization) or criminality (versatility) were more important and found support for the versatility hypothesis. In addition, similar to other crimes, hate crimes are often driven by excitement and thrill seeking (McDevitt et al., 2002). These findings are also consistent with the general theory of crime and point to the need to study whether hate crime offending shares risk factors with other crimes.

Neighborhood disorganization

Social disorganization theory (Shaw and McKay, 1942), with its emphasis on communities and neighborhoods, is the most frequently used macro-level delinquency theory in the United States. In socially disorganized communities, deviant behavior proliferates due to the inadequate reinforcement of prosocial norms and ineffective crime responses (Kornhauser, 1974). Over the last two decades, hate crime research in the United States has focused on communities and found that neighborhood disorder is linked to heightened hate crime rates (Gladfelter et al., 2017; Grattet, 2009; Green et al., 1998; McNeeley and Overstreet, 2018; Mills, 2021). Lyons (2007) suggested that the impact of neighborhood disorder on hate crime varies according to the social group targeted. For example, disorganized areas see more anti-White hate crimes, while organized ones have more anti-Black incidents. These findings are consistent with those of other studies (Gladfelter et al., 2017; Grattet, 2009). In general, hate crimes against minorities often occur in neighborhoods with clear ethnic majorities and growing minority populations. However, these research results are mainly from the United States and may not apply universally (Benier et al., 2016).

Hate crimes also reflect cultural tensions (Van Kesteren, 2016) and intergroup conflicts (Craig, 2002). The ‘defended neighborhoods’ perspective (Grattet, 2009) suggests that these tensions stem from majority groups defending against perceived minority invasions. Green et al. (1999a) found that hate crime perpetrators feared social change more than economic hardship, aligning with GST, which sees noneconomic frustrations as crime risk factors.

Current study

As this brief review of hate crime theorizing suggests, general criminological theories have proved to be fruitful sources for explaining bias-motivated offending, but there remains a need for further examination of the ‘generality’ of hate crimes. The present study also responds to calls by prior researchers to move beyond definitional discussions (Roxell, 2011: 202) and from predominant victim focus to analysis of hate crime offending among adolescents (see, e.g., Hall, 2015; Roxell, 2011). This study further responds to the need to examine cross-national differences in hate crime using comparable individual-level data (e.g., Sheppard et al., 2023). Our analysis focuses on relatively severe physical forms of hate crimes, such as group fights, weapon carrying, assault, and property destruction. In contrast, we do not examine hate-motivated online harassment or online hate, which may be more prevalent among young people.

To our knowledge, the ISRD3 is the only broad and comparative international youth offending and victimization survey to incorporate measures on hate crime perpetration. In addition, ISRD3 contains a wide range of measures on criminologically relevant and theory-based risk factors. We use these valuable data to describe the prevalence of self-reported hate crime offending in school-based city samples of European youth in 17 countries and to compare the individual risk factors associated with hate crime versus nonhate crime offending.

Data

In our analyses, we used the ISRD3–35 dataset as a single sample representing European youths. The ISRD3 is an internationally comparative delinquency survey project targeting youths from Grades 7 to 9 in more than 30 countries (see, e.g., Enzmann et al., 2018). The survey covers a wide range of crime victimization experiences and offending behaviors, along with criminologically relevant correlates and background factors. The data were collected through schools and during school hours. The project will use city samples from one or more cities in each country. These are sampled urban youth populations from the countries and do not represent all adolescents nationally but only those in the selected cities. In all participating countries, the implementation adhered to the ISRD3 protocol (Enzmann et al., 2018). The samples were based on random sampling of school classes in the selected research cities (although random sampling of schools was used in some countries). In Austria and Switzerland, national samples were used. We used the ISRD3–34 version, which became available to the participating researchers in the autumn of 2022.

Follow-up questions on hate offending were present only in the online version of the questionnaire. Therefore, this analysis excludes all ISRD3 countries that used the paper and pencil version of the questionnaire. For the United Kingdom and Germany, it should be noted that the national data collectors used both paper and pencil and online data collection, and this analysis only draws on online parts of those samples. The final sample included 34,871 respondents from 17 European countries and 803 schools.

Measurement of hate crime

The standard ISRD3 online questionnaire incorporated follow-up questions on four types of offending: property destruction, weapon carrying, group fight, and assault. The question sequence logic was the same for all four. For instance, the respondents who admitted to group fights (had ever participated in fights in public places) were asked follow-up questions at the end of the questionnaire, starting with: ‘You said earlier that you have taken part in a group fight in a football stadium, on the street, or another public place. The following questions are about the last time you did this.’

Regarding hate crimes, the respondents were asked whether they had chosen the target based on the victim's group identity. The term hate, or any emotionally charged language, was deliberately avoided. Following conventional usage, we define hate crimes as offenses intentionally targeting specific social groups (see Table 1). The weapon carrying question also included follow-ups about motivation. Instances in which a weapon was carried without intent to harm others (e.g., for reasons related to a hunting or sports hobby) were excluded.

OPEN IN VIEWER

Table 1.

Primary offending questions and follow-ups on biased offense targeting, ISRD3.

Primary offending question	Biased targeting follow-up question
Have you ever in your life…	Did you…
…damaged something on purpose, such as a bus shelter, a window, a car or a seat in the bus or train?	…choose your target because the owner of the property represented some specific group of people?
…carried a weapon, such as a stick, knife, gun, or chain?	…at that time consider some group of people as a threat to you?
…taken part in a group fight in a football stadium, on the street or other public place?	…fight these people because they represented some specific group of people?
…beaten someone up or hurt someone with a stick of knife so badly that the person was injured?	…do it because the person you beat up or hurt represented some specific group of people?

ISRD3: International Self-Report Delinquency Study.

The response options for the hate crime follow-ups included the following categories: ‘no’, ‘immigrants’, ‘people from specific ethnic or national background’, ‘people who speak a language other than my own’, ‘people who have a religion other than my own’, ‘people who have different values and opinions than I’, and ‘sexual minorities’. The respondents could tick several alternatives.

For the analysis, ‘hate crime offenders’ are those whose most recent actions were property destruction, weapon carrying, group fight, or assault targeted at an identity group. ‘Other offenders’ committed at least one of the trigger offenses without identity-based victim selection, while ‘nonoffenders’ had not committed any of them. Nonresponses were coded as missing.

Although the questionnaire covered 13 offenses, only the four trigger offenses were used to classify respondents. As a result, some non-offenders may have committed other crimes. Our analysis focuses on serious physical hate crimes while excluding, for example, online harassment and cyberbullying, which may be more common. Additionally, our measure is conservative, as it omits individuals whose earlier offenses were hate-motivated but whose most recent ones were not. Conversely, measuring hate crime by the most recent offense may include individuals with a history of nonbias offending. Prevalence rates by sampled urban youth populations from all countries and overall are reported in Table 2.

OPEN IN VIEWER

Table 2.

The prevalence of hate crime offending.

Country	Total sample N	Hate crime n	Hate crime prevalence among total sample	Hate crime prevalence among offenders
Armenia	761	15	2.0	12.0
Austria	5046	122	2.4	12.2
Bosnia and Herzegovina	2827	45	1.6	14.0
Estonia	3575	59	1.7	11.0
Finland	2152	26	1.2	6.6
France	1307	22	1.7	6.3
Germany	1278	8	0.6	3.3
Kosovo	1043	19	1.8	23.2
Macedonia	1223	62	5.1	28.7
Netherlands	1785	24	1.3	5.4
Poland	2086	15	0.7	8.4
Portugal	1150	9	0.8	4.6
Republic of Serbia	634	21	3.3	16.9
Spain	1252	18	1.4	9.3
Switzerland	3795	73	1.9	9.1
Ukraine	1625	32	2.0	14.1
United Kingdom	1164	11	0.9	6.4
All	32,703	581	1.8	10.4

Independent variables

In this article, one of our aims is to explore whether variables derived from standard criminological theories are predictors of hate-based victim targeting, as well as of nonbiased motivated offending.

Social strain was measured using three variables. Structural strain was captured by parental employment (0 = both working, 1 = at least one unemployed). Subjective strain was measured by a question asking whether the respondent's family was better off or worse off than other families (scale from 1 = much better off to 7 = much worse off). Family situational strain was measured with a sum variable based on the types of strain: death of a parent, divorce/separation of parents, parental alcohol problems, fights between parents, and serious conflicts between parents (scale 0–6).

Self-control was measured using attitude-based and behavioral measures (see, e.g., Tittle et al., 2003). Attitude-based measurement was conducted using an adapted nine-item version of the Grasmick–Tittle Scale (Grasmick et al., 1993; Marshall and Enzmann, 2012). The adapted version included the dimensions of impulsivity (e.g., ‘I act on the spur of the moment without stopping to think’), risk-taking tendency (e.g., ‘I like to test myself every now and then by doing something a little risky’), and self-centeredness (‘If things I do upset people, it's their problem not mine’) with three items per dimension. Items were summed to a composite variable with a Cronbach's alpha of 0.85 (scale 9–36).

Behavioral measurement of self-control (see, e.g., Hirschi and Gottfredson, 1993; Tittle et al., 2003) was based on the lifetime variety of nonviolent offenses (i.e., self-reported, analogous behaviors; Tittle et al., 2003) (scale 0–9). Only offenses not used as hate crime triggers were included in this measure. Thus, this measure mostly reflects participation in theft offenses.

Perceived neighborhood disorder was captured by a sum variable with five items on perceived crime, drug selling, fighting, abandoned buildings, and graffiti in the neighborhood. The Cronbach's alpha for the measure was 0.86 (scale 5–20).

Social control was measured by a sum variable consisting of five parental supervision-related items, such as ‘If I have been out, my parents ask me what I did, where I went, and who I spent time with’ and ‘If I go out in the evening my parents tell me when I have to be back home by’. The response scale for the items ranged from 1 (never) to 5 (almost always). The Cronbach's alpha for the measure was 0.68 (scale 5–25).

Shame propensity was measured using nine questions. This indicator captures how ‘shamed’ the respondent would feel if he/she was caught shoplifting or physically hurting another person or were arrested by the police for committing a crime and his/her best friend, teacher, or parents found out about it. The Cronbach's alpha for the nine-item measure was 0.89 (scale 9–27).

Routine activities were measured with an item on unsupervised leisure. The respondents were asked how many times a week they usually went out in the evening. This information was coded into a three-level categorical variable (scale 0–7).

Antidiscrimination attitude was measured by a single item asking the respondent to say how wrong he/she considers it to be to ‘knowingly insult someone because of his/her religion, skin color, or ethnic background’ (1 = not wrong at all, 2 = a little wrong, 3 = wrong, and 4 = very wrong).

Hate crime victimization was measured by asking the respondents whether someone had ever threatened them ‘with violence or committed physical violence against you because of your religion, the language you speak, the color of your skin, your social or ethnic background, or for similar reasons?’ (0 = no, 1 = yes).

We used gender, age, and immigrant background (0 = no immigrant background, 1 = second-generation immigrant, 2 = first-generation immigrant) as individual-level control variables in the main theoretical analysis. As a school-level control, we used the aggregated percentage of students with immigrant backgrounds (0 ≤ 15%, 1 = 15–50%, and 2 ≥ 50%) within the school.

For descriptive and multivariate modeling purposes, all (quasi-)continuous variables were transformed into the percent of maximum possible (POMP) scores (Cohen et al., 1999), with lower and upper bounds of 0 and 100. Descriptive statistics of all the independent variables (before standardizing) are presented in Appendix I.

Analysis

In our analyses, we use multilevel (i.e., mixed effect) multinomial and logistic regression modeling to study the risk factors of hate crime offending. We use multilevel multinomial logistic regression to explore the risk factors associated with hate offending and other (nonhate) offending so that both offender types are compared with nonoffenders. In addition to identifying risk factors, the aim of this analysis is to analyze the prevalence of hate crime offending in different European countries while accounting for the studied individual-level factors. Another aim was to specify the risk factors of hate offending and to examine whether hate offending and other (nonhate) offending share the same risk factors. To explore whether some of the shared risk factors may be particularly salient for hate crime, we additionally use multilevel logistic regression within the offender subsample. In that analysis, we ask whether the same factors are related to hate crime when hate offenders are compared with other offenders who have reported one of the four trigger offenses (destruction of property, gang fight, weapon carrying, and/or assault).

Multilevel models were conducted as generalized structural equation models using the GSEM package in Stata 15 statistical software (StataCorp, 2017). To analyze country differences in hate crime offending and to adjust for nation-level variation in our sample designs, we exclude all nation-specific factors by using country dummies in our models (not reported in multilevel regression tables). North Macedonia was treated as a reference country due to its highest hate crime prevalence. In addition, our models adjusted for the clustered error structure at the country level. To further account for the nested structure of our data and the variance associated with school level, our models included random intercepts for schools. Random effects were estimated as unobserved latent variables in the generalized structural equations. For the multilevel multinomial logistic regression analyses, random intercepts were constrained to be equal between the hate offender and other offender models. As recommended by Gelman (2008), (quasi-)continuous variables in our models (transformed into POMP scores; Cohen et al., 1999) have been rescaled to z-values divided by 2 to enable a better comparison of the effect sizes of these variables to the effect sizes of dichotomous variables.

For our models, we report relative risk ratios (RRRs) for the multinomial models and odds ratios (ORs) for the binary logistic regression models, along with the corresponding significance level. For the random part of our models, we report standard deviations for the random intercepts. After the multilevel models, we calculate the prevalence rates of hate crimes among offenders for each country, which are adjusted for the included individual-level variables (Figure 1). Adjusted prevalence rates are based on predicted hate crime offending prevalence while keeping the individual-level variables fixed in their mean values (the logistic model in Table 3). Adjusted prevalence rates were calculated as population average marginal effects (AMEs) (see Figure 1). These AME coefficients can be interpreted as the probability of hate crime offending expressed in percentages; they are reported in Appendix II.

OPEN IN VIEWER

Figure 1.

Prevalence (%) of hate crime offending among 12–16 year olds in 17 European countries adjusted for the individual level variables included in our multivariate models. Panel A: prevalence of hate crime offending among adolescents. Panel B: prevalence of hate crime offending among adolescent crime offenders.

OPEN IN VIEWER

Table 3.

Multilevel multinomial and logistic regression models.

		General offending	Hate-motivated offending	Hate-motivated offending among offenders
Fixed part		RRR	RRR	OR
Parental unemployment (ref. = No)	Yes	1.04	1.15	1.13
Relative deprivation of family		1.00	0.99	1.00
Family problems		1.35***	1.45 ***	1.07
Parental supervision		0.75***	0.76**	1.06
Disorder in neighborhood		1.19***	1.53***	1.40***
Nights out per week		1.68***	2.06***	1.24*
Antidiscrimination attitude		0.90**	0.53***	0.58***
Propensity to feel shame		0.67***	0.62***	0.82*
Self-control		0.58***	0.39***	0.58***
Life-time delinquency		2.38***	2.85***	1.17***
Gender (ref. = Female)	Male	2.08***	2.98***	1.50***
Age		1.01	1.10*	1.09
Hate crime victim (ref. = No)	Yes	1.70***	2.26***	1.36*
Immigrant background (ref. = No)	2nd generation	1.11*	1.52***	1.40*
	1st generation	1.14	1.71**	1.54*
Immigrants in school		1.10	1.21	1.13
Constant		0.22***	0.00***	0.00***
Random part		SD	SD	SD
School		0.34	0.34	0.30
N		32,703	32,703	5588

RRR: relative risk ratio; OR: odds ratio.

Prevalence of hate crime offending among European youths

The overall prevalence of hate crime in the current dataset was 1.8% (see Table 2). The prevalence also varied among urban youth samples from different countries. As shown in Table 2, the highest prevalence was found in North Macedonia (5.1%), while the lowest readings were observed in Germany (0.6%). Among self-reported offenders (i.e., youth who had committed at least one of the four trigger offenses), the prevalence of hate crimes was 10.4%. North Macedonia (28.7%) and Kosovo (23.2%) manifested thehighest percentages, followed by the Republic of Serbia (16.9%), Ukraine (14.1%), and Bosnia and Herzegovina (14.0%) On the other hand, Western European samples appeared to form a cluster of low hate crime motivations. That is, respondents in Germany (3.3%), Portugal (4.6%), the Netherlands (5.4%), France (6.3%),the United Kingdom (6.4%), and Finland (6.6%) showed the lowest hate crime motivation among those who reported involvement in one of the four studied offenses.

Risk factors for hate crime offending

As shown in Table 3, family situational strain was positively associated with hate crime offending (RRR = 1.45, p < .001) and nonhate offending (RRR = 1.35, p < .001). Relative deprivation of the family (or structural strain) or parental unemployment did not statistically significantly predict hate offending or nonhate offending.

Parental supervision was negatively correlated with the likelihood of hate crime offending (RRR = 0.76, p = .007) and nonhate motivated offending (RRR = 0.75, p < .001). Social disorganization as measured by perceived public disorder was a robust risk factor for more likely hate offending (RRR = 1.53, p < .001) and nonhate offending (RRR = 1.19, p = .001). Regarding the leisure time routine activities, the number of nights spent out per week was positively correlated with hate crime (RRR = 2.06, p < .001) and nonhate crime (RRR = 1.68, p < .001).

Antidiscrimination attitude was negatively correlated with a lower likelihood of hate crime offending (RRR = 0.53, p < .001) and nonhate offending (RRR = 0.90, p = .005). Shame propensity was associated with lower odds of hate offending (RRR = 0.62, p < .001) and nonhate offending (RRR = 0.67, p < .001). Attitudinal self-control was negatively associated with hate crime offending (RRR = 0.39, p < .001) as well as nonhate offending (RRR = 0.58, p < .001). In addition, the number of committed nonviolent offenses was positively correlated with hate offending (RRR = 2.85, p < .001) as well as nonhate offending (RRR = 2.38, p < .001). Individuals who had been hate crime victims themselves were more likely to be hate crime offenders (RRR = 2.26, p < .001) as well as nonhate offenders (RRR = 1.70, p < .001) than those without hate crime victimization experience.

Regarding background variables used as controls, males were more likely hate offenders (RRR = 2.98, p < .001) and nonhate offenders (RRR = 2.08, p < .001) than females. In addition, age was positively correlated with the likelihood of hate offending (RRR = 1.10, p = .050). When compared to respondents without an immigration background, second-generation immigrants were more likely to be hate offenders (RRR = 1.52, p = .001) and nonhate offenders (RRR = 1.11, p = .036). First-generation immigrants were also more likely to be hate offenders (RRR = 1.71, p = .002). Interestingly, the proportion of immigrants in school was not associated with hate crime offending or nonhate offending.

When analyzing hate-motivated offending among offenders (Table 3), perceived neighborhood disorder (OR = 1.40, p = .001) and the number of nights spent out per week (OR = 1.24, p = .016) were positively correlated with the likelihood of hate-motivated offending. In addition, antidiscrimination attitude (OR = 0.58, p < .001) and propensity to feel shame (OR = 0.82, p = .045) were associated with a reduced likelihood of hate crime offending.

and individuals with first Attitudinal and behavioral self-control appear to be robust predictors of hate crime offending. Among respondents who reported offending, attitudinal self-control (OR = 0.58, p < .001) and the number of committed nonviolent offenses (OR = 1.17, p < .001) were correlated with hate crime offending. Hate crime offending was also more likely among hate crime victims (OR = 1.36, p = .018), males (OR = 1.50, p < .001), and individuals with second- (OR = 1.40, p = .013) and first-generation immigration backgrounds (OR = 1.54, p = .015).

Adjusted country differences in hate crime offending

To further explore differences in hate crime offending between urban youth populations in 17 European countries, we present the prevalence estimates adjusted for the measured individual-level predictors in Figure 1. The map of countries shown in Figure 1 presents the adjusted hate crime prevalence among adolescents and separately for crime offenders, reflecting the country sample variation in hate crime offending that is independent from the studied individual-level risk factors (see also Appendix II).

According to Figure 1, there are clear international patterns in the prevalence of hate crime offending, with the highest rate in Southeast Europe and the lowest rate in Northern and Western Europe. However, we must keep in mind that our findings reflect comparisons between urban youth populations from these countries (i.e., selected cities) rather than national youth populations. The prevalence of hate crime appears to vary with a distinct pattern across Europe. Countries manifesting a high prevalence of hate crime tend to be in Southeast Europe (i.e., the Balkans and Ukraine), while Western and Northern Europe seems to have lower levels of hate crime among youth. Figure 1 might suggest that youth hate crime offending is related to ethnic, national, and cultural conflicts (e.g., Spain) or proximity to conflict areas. After accounting for individual-level factors, the difference in hate crime offending remained statistically significant between North Macedonia (the reference country with the highest prevalence) and all other countries, except Armenia (OR = 0.97, p = .932) and Kosovo (OR = 0.93, p = .826).

The prevalence of hate crime offending is partly explained by the between-country variation in individual-level risk factors (see Appendix II). Macedonia has the highest hate crime prevalence both before (5.1%) and after adjusting (5.0%) for individual-level risk factors. However, the prevalence of hate crime offending was 1.8% for Serbia after adjusting for the individual-level risk factors, while it was 3.3% before adjustment (the second highest). This suggests that the relatively high hate crime prevalence among Serbian adolescents is partly explained by accumulated risk factors. On the other hand, the prevalence of hate crime offending was the second highest (4.8%) in Kosovo after adjusting for individual-level factors (1.8% before). This, in turn, suggests that the prevalence of hate crimes is higher than expected based on individual-level risk factors in Kosovo, indicating a contextual effect. Similar adjustment effects can be observed in the prevalence of hate crimes among offenders (see Figure 1 and Appendix II).

Contributions and limitations

The ISRD3–35 dataset allows for the exploration of bias-motivated adolescent offending and cross-national comparisons between samples of urban youth populations (i.e., city samples). We document significant country-level differences in self-reported hate crimes and regional patterns. However, the results do not represent young people nationally in these countries but only those in the selected cities included in the study. We also demonstrate the utility of analyzing self-reported foreground factors in identifying hate-motivated offending. Our results suggest that various theoretical perspectives explaining general delinquency—such as strain, social learning, self-control, social disorganization, social bonding, and routine activity—also predict hate crimes. We also identify disorder in neighborhood, risk routines, antidiscrimination attitudes, shame propensity, self-control, lifetime delinquency, and hate crime victimization to be particularly strongly associated with hate crime, as they distinguish hate crime offenders from other offenders (within the offender population). With these results, we contribute to the debate on the universality of crime and advance theoretical understanding (cf. Tittle, 2016). However, causal inferences cannot be drawn due to the observational and cross-sectional nature of our study, and assumed mechanisms require further validation beyond our current analyses.

The limitations of self-report surveys on sensitive topics have been well documented (e.g., Kivivuori, 2014), and the ISRD3 survey is no exception (e.g., Enzmann et al., 2018). Despite its large scale, the sample is still too small to exclusively study hate-motivated offenders. The ISRD3 is a school-based survey, with varying response rates, with a few exceptions (Austria and Switzerland), conducted in two cities across Europe and other parts of the world. Thus, the estimates of prevalence should be approached with caution since these samples are not representative of the youth population in the included countries.

Furthermore, our analyses are restricted to youth admitting to one or more of the four ‘trigger offenses’ (property destruction, group fights, weapon carrying, and assault). While the selected acts cover different forms of hate crime, the indicators used do not include forms such as threats, harassment, bullying, or cyberbullying, which are relevant in the context of hate crime. We also do not capture the most severe forms of hate crimes, such as lethal violence. Future research with a broader range of acts could illuminate the potential similarities and differences between hate-motivated and other types of crime.

Policy implications

This study did not seek to identify causes of hate crime but examined risk factors typical of hate crime offenders, highlighting relevant youth groups for interventions to target. As these risk factors largely overlap with other (violent) crimes, effective general crime prevention will also likely aid hate crime prevention. Our findings suggest that hate crime interventions should specifically focus on youth experiencing neighborhood disorder, engaging in risky routines, holding discriminatory attitudes, exhibiting low self-control and deficient moral emotions, having a delinquent history, and experiencing hate crime victimization.

In conclusion, biased crime is a complex problem that cannot be solved easily. For instance, the finding (Lantz et al., 2024) that hate crime victimization is linked to more discriminatory attitudes suggests the importance of retaliatory violence that may uniquely affect hate crime offending. At the very least, we could aim to interrupt the escalation of youth hate crimes by investing in targeted education programs in immigration-rich schools.