1 The short and longer term impacts of hate crimes experienced directly, indirectly and through the media Jenny L. Paterson1, Rupert Brown2, & Mark A. Walters2 1Teesside University 2University of Sussex Word count: (not including Title page, Tables and Figures) 10652 Running head: Short and long term impacts of hate crime Author note: This research was supported by a Leverhulme Trust Grant (RPG-2013-027) awarded to Rupert Brown and Mark Walters. Correspondence concerning this article should be addressed to Jenny Paterson, School of Social Sciences, Humanities & Law, Teesside University, Middlesbrough, TS1 3BX. Email: j.paterson@tees.ac.uk. 2 Abstract A longitudinal study (N = 774) explored the short and longer-term impacts of anti-LGBT hate crime experienced directly, indirectly and through the media. In the short term, being a victim (direct) or personally knowing of a hate crime victim (indirect) was positively associated with vulnerability, emotional responses and behavioural intentions after reading about a hate crime. Direct victims were also less empathic towards other victims and engaged in more victim-blaming. A structural equation model showed direct experiences (via personal vulnerability and empathy) and media experiences (via group-threat and victim-blaming) to be cross-sectionally associated with behavioural intentions. Media experiences also had lasting demobilising impacts on actual behaviours, again serially mediated by group-threat and victim-blaming. The findings highlight the emotional and behavioural impacts of hate crimes on both direct victims and on the wider LGBT community. They also raise questions about media reporting of hate crimes and the role of victim blaming. Key words: Hate crime; LGBT victimisation; Intergroup emotions; Intergroup threat; Victim blaming. 3 Crimes which are motivated by hostility towards a person’s sexual orientation or gender identity can be devastating for victims (Herek, Cogan, & Gillis, 2002; Walters, Paterson, Brown, & McDonnell, 2017). Individuals may be left with substantial physical injuries (Herek et al., 2002) along with extensive psychological trauma (D’Augelli, Grossman, & Starks, 2006; Walters et al., 2017), all of which may contribute to avoidant behaviours and social isolation (e.g., Perry, 2009). These impacts are also likely to reverberate through communities and individuals who share the targeted characteristic. Research has shown that Lesbian, Gay, Bisexual and Trans (LGBT) individuals who simply know someone who has been a victim of a hate crime (e.g. Perry & Alvi, 2012), or have heard about a hate crime through the media (Noelle, 2002), report similar responses as the direct victims (e.g., vulnerability, anger, anxiety). Since reported hate crimes against LGBT people are on the rise (Bachmann & Gooch, 2017; Corcoran & Smith, 2016), individuals who identify as LGBT are increasingly likely to be a victim, know a victim, or see a media report of an anti-LGBT hate crime. Understanding the impacts of such experiences of hate crimes is, thus, both timely and important. In this study, we explore the impacts of anti-LGBT hate crimes by first comparing how different types of hate crime experiences (e.g., direct and vicarious) impact on individuals’ perceptions, feelings, and behaviours. Second, we explore how these experiences lead to such responses in the short-term. Finally, utilising a longitudinal design, we investigate the longer-term impacts of hate crimes. In doing so, we aim to make significant contributions to the hate crime, intergroup relations, and victim blaming literatures, as well as providing data that can help inform current criminal justice debates surrounding the impacts of targeted victimisation of LGBT people and the support of victims (e.g., Walters et al., 2017). 4 Hate Crimes Experienced Directly, Indirectly and via the Media Most research investigating the impacts of hate crime has examined the experiences of direct victims – individuals who are singled out and abused by perpetrators (e.g., Herek et al., 2002). Studies have shown that anti-LGBT offences are more likely to involve physical violence than parallel non-hate crimes (Cheng, Ickes, & Kenworthy, 2013; Herek et al., 2002), often resulting in direct victims exhibiting acute signs of psychological trauma, including PTSD (e.g., D’Augelli et al., 2006). Victims are likely to feel vulnerable to further attacks, anger at their victimisation and may consequently alter their behaviour, for example by avoiding certain locations or hiding their sexuality or gender identity (Herek, Gillis, & Cogan, 1999). It is not only direct victims who may experience such harms. Hate crimes are symbolic attacks against “difference” which are intended to send messages of hate to entire groups of people. A growing body of research has begun to quantify the vicarious experiences of hate crimes (Bell & Perry, 2015; Iganski & Lagou, 2015; Noelle, 2002; Perry & Alvi, 2012; Walters et al., 2017). Within this nascent literature, these vicarious experiences of hate crime have been conceptualised in different ways. Noelle (2002), for example, showed that the media coverage of the homophobic murder of Mathew Shepard (in 1998) impacted the world-views of other LGB people. Others have shown that personally knowing other hate crime victims in their social network has adverse consequences (Walters et al., 2017). Though both approaches are similar, we believe that they represent two different types of vicarious hate crime victimisation: the former is via the media and the latter, more personal, experience is what we refer to as indirect victimisation. Regardless of the type of vicarious experience, both are likely to affect other group members. First, for those who share the targeted identity (i.e., other LGBT people), simply hearing of others’ victimisation is likely to make them feel personally vulnerable – that they 5 could be next (Paterson, Brown, & Walters, 2018). Second, in line with social identity approaches (e.g., intergroup emotions theory (IET), Mackie, Devos, & Smith, 2000; social identity theory (SIT), Tajfel & Turner, 1979), people who share the characteristic are also likely to view the hate crime as an attack on the group as a whole. Accordingly, individuals are more likely to respond as a group member and not just as a vulnerable individual (e.g., Smith, 1993). This group-based response is then likely to motivate ingroup members to react more strongly in order to protect themselves, their group’s status and rights, and the security of their fellow group members (Paterson, Brown, & Walters, in press). Although these findings all support the assumption that hate crimes have negative effects on individuals and communities, there has yet to be a study that includes measures of all three experiences simultaneously. Such an investigation is important as it would highlight the possible risks associated with each type of victimisation. For example, on learning of a new hate crime, while those who have experienced direct victimisation – due to its more personal nature – might be expected to experience more pronounced effects, those who have previously suffered vicarious forms of victimisation (indirect and via the media) would be expected to show report greater impacts than those who have had no such prior experiences (Hypothesis 1). The Pathways of Hate Below, we draw upon the existing literature to outline likely mechanisms through which the different types of hate crimes may impact LGBT individuals’ feelings of vulnerability, perceptions of hate crime victims, emotional reactions to hate crimes, and behavioural responses to hate crimes (Figure 1). 6 Step 1. Experiences of Hate Crimes and Perceptions of Threat As hate crimes target characteristics that are generally stable (i.e., sexual orientation and/or gender identity: Savin-Williams & Ream, 2007), direct victims, due to their likely marginalised status, often continue to feel vulnerable to further hate-motivated attacks (Herek et al., 1999). Furthermore, they may view the attack on their LGBT identity as an attack on the “LGBT identity” as a whole, thus highlighting hate crimes to pose a realistic, physical threat to other LGBT people, as well as a symbolic threat against LGBT rights, culture and way of life (see Cottrell & Neuberg, 2005 on the distinction between realistic and symbolic threats). Direct experiences of hate crime, then, are likely to be associated with increased feelings of personal vulnerability and perceptions of group-based threat (Hypothesis 2a). Similarly, vicarious experiences of hate crime are likely to make other LGBT individuals’ feel that their identity group is under threat and that they, as a group member, are vulnerable too. However, research has suggested that some group members minimise their feelings of personal vulnerability after hearing of a hate crime by distancing themselves from the victim and believing that they would not have been victimised as they would have avoided certain locations or hidden their sexual orientation (Noelle, 2002). Thus, vicarious experiences are likely to increase perceptions of threat posed to the group in general, but the literature is unclear whether these experiences would also lead to feelings of personal vulnerability (Hypothesis 2b). Step 2. Perceptions of Threat and Perceptions of Other Victims The experience of hate crimes and the resulting feelings of vulnerability and threat are likely to impact on how individuals respond to hearing about others’ hate-based victimisation, including how much empathy they feel for the victim(s) and to what extent they blame the victim(s). Cortland and colleagues (2017) recently found that sharing similar 7 discriminatory experiences, akin to hate crimes, promoted more empathy for other victimised groups. In one of their studies, Black people were more likely to be empathic towards gay people when discrimination (equal marriage) was framed as a ‘civil rights issue’ compared to a ‘gay issue’. This suggests that the ‘civil rights’ terminology increased the saliency of the vulnerability of Black people and so resulted in them being more empathic towards the discrimination of gay people. Thus, when LGBT people feel vulnerable or that their group is at risk, they will be more empathic towards ingroup victims of hate crimes (Hypothesis 3a). In addition, LGBT individuals who feel more vulnerable and threatened may be less likely to blame hate crime victims because they understand and appreciate the hostility that LGBT people face is pervasive and usually unavoidable. Nevertheless, research has also shown that victim blaming may be higher for those most vulnerable to similar hate-motivated crimes (Bell & Perry, 2015). A gay man participating in Noelle’s (2002) qualitative study, for example, insinuated Mathew Shepard may have had a role to play in his tragic murder by remarking, “I wouldn’t personally even go into a bar, in Wyoming, where there was a mixed crowd, and then be shooting off my mouth that I’m gay” (p. 40). Such statements reveal an element of victim-blaming that indicates an attempt to exert control over one’s own perceived risk of future victimisation – if I don’t act openly gay, I won’t be attacked. In addition, this tendency to victim blame is likely to be a product of the heteronormative societies in which LGBT people are ostracised for threatening sexual and gender norms (Bibbings, 2004). For instance, individuals who have had to “come out”, do so because there remains a general environment that positions heterosexuality as the dominant norm. Sexual orientations that subvert this norm have resulted in hostilities towards LGB people, including parents who disapprove of homosexual relationships. This puts additional emotional burdens on LGB people to conform to a heterosexual lifestyle (including “acting straight”) if they are to “fit in” to society. Surrounded by such negative perceptions can lead to LGBT people 8 internalising societal homophobia and genderism and, consequently, believing that LGBT people who subvert sexual or gender norms are (partly) to blame for any abusive reactions to their sexual orientation (Herek, Cogan, Gillis, & Glunt, 1998). Together, this suggests that, while feelings of vulnerability and group-based threats are likely to be positively associated with empathy for the victim (H3a), such feelings may also increase negative reactions to the victims in which they are (partly) blamed for their victimisation (Hypothesis 3b) Step 3. Perceptions of Victims Influence Emotional Reactions to Hate Crimes Empathy is generally defined as the process in which individuals recognise and match the emotions of someone in distress and, in turn, are motivated to act altruistically (Batson & Ahmed, 2009). As feelings of anger and anxiety are common in direct victims (Herek et al., 1999), it is likely that those who feel empathy for the victim will match such emotions. Indeed, previous research has shown such a link: empathy for the victim is positively associated with feelings of anger and anxiety towards hate crime (Hypothesis 4a: Paterson et al., 2018, in press) Victim blaming, meanwhile, is also likely to impact on the emotions felt towards the hate crime. As noted, group members can blame the victim when they feel particularly vulnerable to hate crimes and because of internalised homophobia (H3b). As internalised homophobia already causes LGBT people to feel ashamed of themselves and their identity group (Herek et al., 1998), personally engaging in such victim blaming reinforces heteronormative ideals and is, thus, likely to compound these feelings of shame about themselves and their group. Consistent with this, Tangney and colleagues (1990; 1992) have consistently found positive associations between blame and shame (but not between blame and guilt). Such findings lend support to our hypothesis that victim blame will be positively associated with feelings of shame towards the hate crime (Hypothesis 4b). 9 Step 4. Emotional Reactions Predict Behaviours IET is well placed to explain the emotional and behavioural impacts of hate crimes (Mackie, Smith, & Ray, 2008; Smith, 1993). This theory suggests that, when group identities are salient, specific group-based emotions will motivate relevant behaviours to help the group. For example, previous research has shown that group-based anger leads to more approach related behaviours, including confrontation (Iyer & Leach, 2008; Mackie et al., 2000) and collective pro-action, perhaps in an effort to re-establish a semblance of power and control over the situation (Iyer, Schmader, & Lickel, 2007; Leonard, Moons, Mackie, & Smith, 2011). Intergroup anxiety, conversely, is thought to lead to more avoidant behaviours which helps put individuals and their group out of harm’s way (Stephan, 2014; Stephan & Stephan, 1985). Likewise, group-based shame has been implicated in avoidant strategies (Sheikh, 2014), but is also thought to lead to a “hostile, humiliated fury” which motivates the shamed individual to retaliate against the (real or imagined) protagonist (Tangney et al., 1992, p. 670). Drawing on this research, then, suggests that once hate crimes elicit emotions (through empathy and blame: H3a and H3b), these group-based emotions will trigger specific behavioural reactions. Anger will be positively related to approach behaviours (Hypothesis 5a); anxiety will be positively associated with avoidant behaviours (Hypothesis 5b), and shame will be positively associated with avoidant and retaliatory behaviours (Hypothesis 5c). Step 5. Behavioural Intentions Translate to Subsequent Behaviours The hate crime literature has shown that victims of direct and vicarious forms of hate crime may change their behaviour as a consequence of their victimisation. Some become socially withdrawn, while others become engaged in more proactive responses (Perry, 2009; Perry & Alvi, 2012). Though this previous research is valuable in understanding the potential behavioural effects of hate crimes, the cross-sectional nature of the research has relied on 10 participants’ predictions of their behaviours (Walters et al., 2017) or participants’ current assessments of their behaviours (Perry & Alvi, 2012). Such designs are limited since participants may not be able to accurately predict their subsequent behaviours (though cf. Ajzen, 1991 for a counter argument). Here we add a longitudinal component to address these limitations. By asking respondents what they intend to do and, at a later date, obtaining reports of what they actually did, we are not only able to understand the effects of hate crime experiences on individuals’ current intentions to act, we are able to show how experiences affect actual behaviours in the longer term. A longitudinal design also permits slightly stronger causal inferences. By testing reciprocal statistical associations, one can determine if a variable at time 1 (e.g., experiences) predicts a variable at time 2 (behaviours) or vice-versa (Kelloway, 2014). Thus, this analysis can help to answer whether experiences with hate crimes impact subsequent behaviours, or whether these behaviours also affect future experiences. Overview and Hypotheses This study used a two-wave longitudinal design over a three-month time period to understand the short and longer term impacts of direct and vicarious experiences of anti- LGBT hate crimes in the UK. At both time points, we assessed participants’ previous experiences with hate crimes (direct, indirect, media) and their current perceptions of general threat and personal vulnerability to hate crime. We then assessed their reactions to a specific hate crime reported in the media, including their reactions to the victim (empathy and blame), their emotional reactions to the crime (anger, anxiety and shame), and their intended behavioural responses (avoidance, pro-action, retaliation, increased security, and raising others’ awareness). We also assessed participants’ self-reported behaviours between Time 1 and Time 2. We tested the following hypotheses: 11 • H1: The impacts of direct experiences of hate crimes will be more pronounced than the impacts associated with indirect experiences of hate crimes which, in turn, will be associated with stronger impacts than having no experiences of hate crimes. • Direct experiences will be positively associated with feelings of personal vulnerability and perceptions of group threat (H2a) while vicarious experiences will be associated with perceptions of threat against the group (H2b). • Feelings of vulnerability and group-based threat will be associated with more empathy for the victim of a hate crime (H3a) and more blame attributions (H3b). • Empathy for the victim will be positively associated with feelings of anger and anxiety towards the crime (H4a), while blame attributions will be positively linked to feelings of shame (H4b). • Anger towards the crime will be positively associated with approach behaviours (i.e., raising awareness, pro-action, retaliation, less avoidance, H5a); anxiety will be positively related to avoidant behaviours (i.e., avoidance, security measures, H5b); and shame will be positively correlated with avoidant behaviours and retaliatory wishes, H5c. • H6: Behavioural intentions at Time 1 will predict actual behaviours reported at Time 2. Method Participants Initially, 835 participants who identified as LGBT and resided in the UK were recruited by Qualtrics to complete a study about ‘Hate Crime Experiences’. Participants were informed that they were to complete the study at two-time points three months apart (February – May 2016) in exchange for a reward offered by Qualtrics (e.g., money, 12 vouchers). Of these 835 participants, 774 completed the first phase and a total of 465 participants completed both phases (retention rate 60%). Here we present demographic information for the final sample: a range of genders were reported - 286 males, 162 females, 12 trans, 1 agender, 1 male and female, 1 trans male, 1 trans female, and 1 intersex; and a range of sexual orientations were reported - 212 gay, 159 bisexual, 51 lesbian, 15 asexual, 14 straight, 10 queer and 4 pansexual. The majority of participants were White British (403), 19 White Other, 11 Multiple/mixed ethnic groups, 10 Black British, 7 Indian, 4 Pakistani, 3 Black Other, 2 New Zealand/Australian, 1 European and 1 Chinese1. The average age at Time 2 was 42.08 years, SD = 14.86. Measures All measures used a 1 (Strongly disagree) to 7 (Strongly agree) point Likert scale and were measured identically at Time 1 and Time 2, unless otherwise stated. Past experiences of anti-LGBT hate crimes and hate incidents2 were defined using the College of Policing operational guidance definition (2014, p. 4) “Any criminal offence, or non-crime incident, which is perceived, by the victim or any other person, to be motivated by a hostility or prejudice based on a person’s sexual orientation or perceived sexual orientation or transgender identity or perceived transgender identity.” Participants then indicated how many times they had been “verbally abused (e.g., shouted at, called names)”, “abused online (e.g., on Twitter, Facebook, discussion boards)” and “physically assaulted (e.g., pushed, punched, kicked)” “by a person who you think was motivated by a prejudice against LGBT people?” The time frame for Time 1 was the past three years and the timeframe for Time 2 was since the last study (approximately 3 months). Response options for verbal and online abuse were: 0, 1-2, 3-4, 5-6, 7-8, 9-10, and 11 times or more, while the response options for physical assaults were 0, 1, 2, 3, 4, 5, and 6 times or 13 more. However, as responses showed significant skew, these measures were dichotomised to categorise people as having been victimised or not in each of the three ways (verbal, physical, online). We then averaged these three dichotomous values to obtain a measure of direct experiences: Time 1 α = .81, Time 2 α = .84. Severity of their worst experience was assessed using the item “Thinking about the time(s) when you have been a victim of a hate crime or incident, how severe would you rate your worst experience?” (1 =Very minor to 7 = Very severe with a “N/A I've not been a victim” option). Indirect experiences with hate crimes and severity of their indirect experience were assessed using the same items but referred to how many people they knew who had been victims (they did not need to be close friends with the victim(s) but did need to know them, i.e., not just seen on TV). Again, these experiences showed significant skew and so were dichotomised (Time 1 α = .80, Time 2 α = .82). Media experiences of hate crimes were assessed using the item “How many instances of anti-LGBT hate crimes and incidents would you estimate that you have seen or read about in the media (e.g., TV, radio, Facebook, Twitter, etc.)?” Response options were 0, 1-10, 11- 20, 21-30, and 31 instances or more. The severity of media experiences was assessed in a similar fashion to the direct and indirect experiences. General perceptions of hate crime. Personal vulnerability to hate crimes was measured using three items: “I worry about being a victim of an anti-LGBT hate crime or incident”, “Being LGBT makes me feel vulnerable to anti-LGBT hate crimes and incidents”, and “Showing affection to my partner in public makes me feel vulnerable to anti-LGBT hate crimes” (Time 1 α = .86, Time 2 α = .88). Two reversed items “I do NOT feel that anti-LGBT hate crimes and incidents represent an immediate threat to me” and “I feel safe from anti- 14 LGBT hate crimes and incidents in the area I live in” were dropped as the factor analyses for both time points revealed that these items loaded on to a separate factor3. Realistic threat to LGBT people was measured using five items adapted from Cottrell and Neuberg (2005) with the stem “I believe anti-LGBT hate crimes and incidents…” “Pose a physical threat to LGBT people”, “Endanger the safety of LGBT people”, “Pose a threat to the possessions of LGBT people”, “Pose a threat to the emotional well-being of LGBT people”, and “Pose a threat to the personal rights of LGBT people”, Time 1 α = .95, Time 2, α = .95. Symbolic threat used the same stem and the following three items: “Pose a threat to LGBT culture”, “Pose a threat to LGBT people's way of life”, and “Pose a threat to the beliefs and values of LGBT people”, Time 1 α = .94, Time 2 α = 95. Reactions to a specific hate crime incident. Participants then read an article described as having recently been published on a British news website with the headline “Gay man beaten in homophobic attack”. Following the article and using a 1 = ‘Not at all’ to 7 = ‘Very much so’ scale, participants were instructed “Thinking about the victim, to what extent do you….” and then indicated their empathy for the victim using four items: “Empathise with the victim”, “Feel sadness for the victim?”, “Feel sympathy for the victim?”, and “Feel respect for the victim?” (Time 1 α = .91, Time 2 α = .92). Participants then indicated to what extent they blamed the victim for the attack using three items, “Think the victim was at fault?”, “Think the victim was irresponsible?” and “Think the victim was reckless?” (Time 1 α =.96, Time 2 α = .97). Emotions towards the article were again assessed on a 1 = ‘Not at all’ to 7 ‘Very much so’ scale with the stem question: “How does reading the article make you feel?” Anger was then measured using four items: “Angry”, “Outraged”, “Annoyed” and “Appalled” 15 (Time 1 α =.91, Time 2 α = .92). Anxiety was measured using two items: “Afraid” and “Anxious” (Time 1 r = .78, Time 2 r = .80) and had initially included “Alarmed” but modification indices suggested that this item also cross-loaded onto anger (χ2(1)∆ = 77.89, p < .001) and so was dropped. Shame was initially measured using three items: “Ashamed”, “Embarrassed”, and “Guilty”, but since these three items are likely to be conceptually distinct (see Tangney, 1990 for an overview), we decided to include only the face valid item of “ashamed” in the analysis. Behavioural intentions were then measured with the lead “Having read the article, to what extent would you consider…” Avoidance was then measured using three items “Going out less often”, “Seeing friends less often”, and “Avoiding certain places and people” (Time 1 α =.88, Time 2 α = .88). Security was measured using two items: “Enrolling in a self-defence class” and “Improving the security of my home and my personal belongings (e.g., change locks, change passwords, improve house alarms)”, Time 1 r = .74, Time 2 r = .75. Pro-action was measured using three items: “Joining and/or increasing my participation in anti-hate crime groups”, “Joining and/or increasing my participation in groups and charities that help LGBT people”, and “Becoming more actively involved with groups that advocate LGBT rights” (Time 1 α =.94, Time 2 α = .95). Raising others’ awareness was measured using two items: “Using social media (e.g., Twitter) to raise others’ awareness of the crime” and “Telling other people (e.g., family and friends) about the crime” Time 1 r = .66, Time 2 r = .65. Initially “Naming and shaming the attackers on social media, if I could” was also included in the scales but since the item was found to cross-load onto actual retaliation (see below), the item was removed from these scales for consistency. Retaliation was measured using two items: “Verbally abusing the offenders if I had the opportunity”, and “Physically attacking the offenders if I had the opportunity”. As these latter two items evidenced significant skew, we created a retaliation scale to indicate those who were against retaliation 16 of any sort (scoring < 4 on both scales) and those who were not entirely against it (scoring 4 or more on either scale). Participants also indicated the strength of their LGBT identity using four items: “I feel good about being an LGBT individual”, “I identify with other LGBT people”, “I am like other LGBT people” and “Being LGBT is an important reflection of who I am” (Time 1 α =.85, Time 2 α = .87). At Time 2, participants completed an additional five items that measured self-reported actual behaviours since Time 1. Using the stem “We would now like to know about whether your behaviour has changed in any way since completing the previous survey. Thinking about the past 3 months, please indicate how often you have done the following – compared to what is usual for you” (1 = less often than usual, 4 = same as usual, 7 = more often than usual) and then participants completed the same items as the behavioural intentions above: Avoidance, α =.65; Security r = .68; Pro-action, α =.95; Retaliation, r = .92. Modification indices suggested that the item “Named and shamed the attackers on social media” cross- loaded onto retaliation (χ2(1)∆ = 173.27, p < .001) and so was dropped from the Awareness scale (r = .64). Results Analytical Strategy We first analysed participant attrition across the two-time points, following which we tested our hypotheses by splitting the analysis into three sections. First, exploiting the larger sample size and therefore greater power of Time 1, we conducted a MANCOVA to compare the effects of having direct, indirect and no experiences of hate crimes (H1). We then tested a longitudinal structural equation model (SEM) to test how experiences with hate crimes impacted on participants’ behavioural intentions at Time 1 and actual behaviours at Time 2 17 through the hypothesised mechanisms (H2 – H5: Figure 1). We then supplemented these analyses with a cross-lagged SEM to investigate the temporal relationships between experiences of hate crimes and behavioural intentions (H6). Attrition As noted earlier, we experienced a 40% attrition rate. Using chi square tests and a MANOVA t-test, we found that those who completed the studies at both time points (‘Completes’ n = 465) differed significantly from those who only completed Time 1 on several variables (‘Drop-outs’ n = 309). Demographically, the Drop-outs were younger (34.50 years vs. 41.53 years, F(1,772) = 45.54, p < .001, ηp 2 = .05), were more likely to be female (42.4% vs. 35.3%: χ2(1) = 4.00, p = .046) and less likely to be male (50.5% vs. 61.1%: χ2(1) = 8.49, p = .004). There were no differences in the number of trans people in the groups or intersex people (both χ2s(1) < 3.45, ps > .063). The samples also differed in the participants’ sexual orientation (χ2(11) = 42.48, p < .001: with more bisexual and straight people and fewer gay people in the Drop-out group) and ethnicity (χ2(13) = 39.45, p < .001: with fewer White British people in the Drop-out group). The two groups also differed in their experiences and reactions to hate crimes. Drop- outs were more likely to have been a direct and indirect victim of all types of hate crimes (all χ2s (1) > 11.51, ps < .001) and these experiences were rated as more severe (Fs(1,772) > 9.37, ps < .002, ηp 2s > .01). MANOVAs also revealed that Drop-outs reported viewing more media reports of hate crimes and showed higher scores on vulnerability, victim blame, anxiety, shame, avoidance, security, pro-action, awareness, and retaliation, all Fs(1,772) > 5.03, all ps < .025, ηp 2s > .01. There were no differences between the groups on the severity of media experiences, realistic or symbolic threat, empathy, anger, or LGBT identity. 18 Despite these compositional differences indicating that some of the attrition may have been non-random, comparisons of the correlation matrices of the two groups indicated that the associations between the variables were very comparable: 98% of the 171 correlations were in the same direction, with 92% of them being the same significance status (i.e., either significant or non-significant in both matrices). Only three correlations changed sign, and all three were non-significantly different from zero. Such similar pattern of responses suggests that the results presented below are unlikely to have been affected by participant attrition. Indeed, in light of the fact that the participants who dropped out had more experiences of hate crimes and stronger reactions, it suggests that the absence of these participants from the ‘complete’ sample makes our longitudinal results a more conservative estimate of the impacts of hate crime. Impacts of Direct, Indirect and No Experiences of Hate Crimes We first compared the impacts of those who had only indirect experiences of hate motivated physical assaults (‘Indirect only’ group: N = 219; Mage = 36.34, SD = 14.99) to those who had no experience (“No experience’ group: N = 392; Mage = 42.93, SD = 15.28) and those who had both direct and indirect experiences5 (‘Both’ group: N = 163; Mage = 31.82, SD = 9.75). As the average age of participants in the three groups significantly differed from one another (F(2,770) = 39.42, p < .001, ηp 2 = .09), a MANCOVA, controlling for age, was conducted on all measures and the findings are presented in Figure 26. The MANCOVA revealed that the groups significantly differed on all variables (Pilai’s Trace = .40, F(28,1514) = 13.34, p < .001, ηp 2 = .20; all univariate Fs(2,769) > 4.17, ps < .016, ηp 2s > .01). As hypothesized (H1), having both direct and indirect experiences (‘Both’ group) was generally associated with more pronounced reactions to hate crimes, especially compared to participants who had no experiences. Victims of both experiences felt 19 more personally vulnerable to hate crimes, more anxious and more shame in relation to the hate crime article, and were more likely to indicate that they would engage in all the behaviours assessed (all ps < .001). They were also less empathic towards the victim in the article and engaged in more victim blaming than participants who had not been victimised (both ps < .001). In partial support for H1, participants who had both experiences indicated that they were significantly more likely than participants who had only indirect experiences to engage in four of the five behaviours (all ps < .015, except raising others’ awareness, p = .440) and reported more anxiety and shame (both ps < .001). Nevertheless, contrary to H1, compared with having only indirect experiences, those participants who had both experiences rated their indirect experiences as less severe (p = .017), showed less empathy and more victim blaming (both ps < .001). They also felt that their group was under less threat physically (realistic threat, p = .002) and felt less angry in response to the hate crime article (p = .009). The impacts of indirect experiences of hate crime were more consistent. Compared with people who had had no experience of hate crimes, and consistent with H1, those who knew of someone else who had been assaulted in the past three years felt significantly more vulnerable towards hate crimes (p < .001) and thought hate crimes pose a more realistic (p < .001) and symbolic threat to LGBT people (p = .008). They also reported significantly stronger reactions to the hate crime reported in the article: they were more empathic (p = .010), angry (p < .001), and anxious (p < .001), and they engaged in marginally less victim blaming (p = .055). They were also more likely to engage in four out of the five behavioural intentions (all ps < .016; but retaliation, p = 1.00) and were more strongly identified as LGBT (p = .028). Taken together, these findings highlight the powerful indirect impacts of hate crimes. They also suggest that, while direct victimisation may enhance some responses in the 20 predicted direction, its effects, particularly on the perception of other victims, may not be as predictable. Structural Equation Model Next, we tested our hypothesised model in which previous experiences affected current thoughts and feelings which led to future actions (Figure 1). Table 1 presents the means, standard deviations, and correlations of the variables used in the SEM (n = 465). In general, participants did not feel especially personally vulnerable to hate crimes but did believe that hate crimes posed a realistic and symbolic threat to LGBT people as a group. After reading the hate crime article, participants showed empathy to the victim and did not blame the victim for the crime. The article elicited more anger than anxiety, and shame was the least pronounced reaction. Participants were most likely to engage in spreading the awareness of hate crimes, followed by joining relevant LGBT groups, improving security and engaging in avoidance, while 36% indicated that they were not wholly against retaliation. Three months later, participants reported that their behaviours had remained relatively unchanged with all five behaviours hovering around the midpoint of the scale (Ms = 3.69 – 4.10). We used Mplus to test our model and, as some variables were non-normally distributed, the MLR estimator was used (Muthén & Muthén, 2011). Factors at the same level were allowed to covary, and all factors were predicted to impact on actual behaviours at Time 2. Age was used as a covariate. All items loaded significantly on to their relevant factors (p < .001). Two indicators of the realistic threat scale were allowed to covary as were two indicators of the empathy scale, and actual avoidance scale. The absolute fit indices of the model were good: RMSEA = .047 [95% CIs: .044/.049], SRMR = .087 and the comparative fit index was adequate (CFI = .921), especially considering the complexity and temporal 21 nature of some of the relationships (Hu & Bentler, 1999). The chi-square was significant, as expected in a large model: χ2(792) = 1719.366, p < .001 (MLR scaling factor = 1.114) (Kelloway, 2014). Figure 3 presents the significant correlational paths and shows general support for the majority of the hypotheses. Direct experiences were positively correlated with feelings of personal vulnerability (H2a) and experiences via the media were positively related to perceptions of group threat, although indirect experiences were not, once direct and media experiences were controlled for (H2b). In addition to the hypothesised relationships, media experiences were positively related to personal vulnerability, while, contrary to H2a, direct experiences were negatively associated with realistic group threat. All three threats were then positively related to empathy for the hate crime victim as predicted (H3a), and vulnerability and symbolic group threat were positively associated with blame, as expected (H3b). However, contrary to H3b, realistic group threat was negatively associated with victim blaming. In support of H4a and H4b, empathy was positively associated with anger and anxiety, while blame was positively associated with shame. In addition, victim blaming was found to be positively related to anxiety. In the next step of the model, anger and shame were related to behavioural intentions in the expected fashion. Anger was positively associated with awareness, pro-action, retaliation, and was negatively associated with avoidance (H5a), and shame was positively associated with avoidance, security and retaliation (H5c). Anxiety was found to be positively linked to avoidant behavioural intentions (e.g., avoidance and security) as expected (H5b), but was also positively associated with the other behavioural intentions: awareness, pro-action, and retaliation. In the last step of the model, H6 received partial support. Three out of the five behavioural intentions were positively related to their actual behaviours at Time 2 (awareness, pro-action, and retaliation). Interestingly, all actual 22 behaviours were positively and significantly predicted by victim blame, even when controlling for all the other factors. Mediational Analyses The ‘model indirect’ command using 1000 bootstrap resamples and 95% confidence intervals was used in Mplus (Muthén & Muthén, 2011) to explore the indirect effects of hate crime experiences on shorter term behaviours (intended behaviours at Time 1: Table 2) and longer term behaviours (actual behaviours at Time 2: Table 3). Table 2 shows that direct experiences of hate crimes indirectly influenced four of the five behavioural intentions at Time 1 (all but retaliation)7. The significant mediational pathways revealed that direct experiences were associated with stronger feelings of vulnerability (H2a) which were, in turn, associated with more empathy for the victim (H3a). This empathy was then related to increased anger and anxiety (H4a). Anger was then associated with the more approach behaviours (more pro-action, more awareness, and less avoidance but no mediation to retaliation: H5a), while anxiety was associated with avoidant and security related behaviours (H5b), as well as more pro-action. Table 2 shows that media experiences indirectly influenced all five behavioural intentions at Time 1 and did so through a variety of paths. First, media experiences were positively associated with perceiving hate crimes to pose a realistic threat to LGBT people (H2b). This perception of threat, in turn, was positively associated to empathy (H3a) and these feelings of empathy were, as stated above, associated with feelings of anger and anxiety (H4a, H4b). Anger on this path was, again, associated with less avoidance and more pro- action and awareness raising but not retaliation (H5a). Likewise, anxiety was positively associated with avoidance and security (H5b). Anxiety, on this path, also increased both pro- action and awareness. 23 In addition to the empathic pathways to behavioural intentions, the effect of media experiences on behavioural intentions was mediated through victim blame. As above, media experiences were positively associated with perceptions of realistic group threat (H2b). Contrary to H3b, this threat was negatively associated with victim blame. Blame was then positively associated with shame as expected (H4b) and also with anxiety. Shame was then positively correlated with avoidance, security and retaliation. The anxiety pathway, meanwhile, positively predicted all five intentions. Using the same mediational analyses as above, we next examined the longitudinal mediational pathways of how hate crime experiences may influence actual behaviours three months later. Table 3 displays the total, direct, and indirect effects along with the significant pathways. It is important to note here that, due to potential suppression effects in which direct and indirect effects are opposite in sign making the total effect non-significant (MacKinnon, Krull, & Lockwood, 2000), we explored the significance of all potential pathways even if the overall effect was non-significant. We found that only media experiences of hate crime were significantly related to actual behaviours at Time 2, and these effects were all indirect. Media experiences were related to more awareness raising behaviours, sequentially mediated through realistic threat, empathy, anger and the intention to raise awareness. The experiences were also related to less pro-action, sequentially mediated by realistic threat, less blame, anxiety, and intentions to engage in pro-action. Notably, in addition to these two mediational pathways, media experiences indirectly – and negatively – affected all behaviours at Time 2. Media experiences were positively associated with perceptions of realistic group threat which were, in turn, negatively associated with victim blame, which was then subsequently positively related to all measured behaviours at Time 2. These findings suggest that media experiences have an overall negative impact on actual behaviours because, while media 24 accounts may increase perceptions of group threat, this threat decreases blame attributions which, in turn, decreases the likelihood of engaging in any subsequent behaviours. Supplementary Analyses To understand the temporal relationship between experiences of hate crimes and behaviours, we conducted a cross-lagged SEM in which experiences (direct, indirect, media) at Time 1 were estimated to predict behavioural intentions (avoidance, security, pro-action, awareness, retaliation) at Time 2, controlling for behavioural intentions at Time 1. Simultaneously, behavioural intentions at Time 1 were estimated to predict experiences at Time 2, controlling for experiences at Time 1. Factors on the same level were covaried (e.g., direct with indirect and media), as were items that were used at both time points (e.g., ‘direct’ at Time 1 with ‘direct’ at Time 2 to account for method covariance, Kelloway, 2014). Age was again used as a covariate and the model fit the data well: χ2(506) = 982.532, p < .001 (MLR scaling factor = 1.113), CFI = .955, RMSEA = .045 [95% CIs: .041/.049]; SRMR = .043 (Hu & Bentler, 1999). Figure 4 shows that direct experiences at Time 1 predicted pro- action, awareness and avoidance at Time 2, while avoidant behavioural intentions at Time 1 predicted greater experiences via the media at Time 2. Discussion Overall, our research clearly shows that anti-LGBT hate crimes, whether experienced directly, indirectly or through the media, have substantial short and longer-term impacts on LGBT individuals and communities. Short-term impacts are well explained by our expansion of IET since they show that the emotions elicited by threat and vulnerability seem to be generated by empathy and victim blaming. In a unique contribution to the hate crime literature, we also provide important longitudinal evidence, suggesting that hate crimes seen 25 through the media have durable – and negative – effects on actual behavioural engagement in response to hate crimes. Substantiating previous research, our results show that anti-LGBT hate crimes have significant impacts on direct victims (Herek et al., 1999). Compared with people who had no experiences of hate crime, direct victims were found to feel more vulnerable to future attacks and were more likely to change their behaviours in response to a reported hate crime. Our study also showed that hate incidents are acutely felt throughout targeted communities (H1). Simply knowing someone who had been a victim of a hate crime impacted on other LGBT individuals’ perceptions of threat and their emotional and behavioural responses to hate crimes, in a manner similar to direct victims. Although having only indirect experiences seemed to promote more positive and sympathetic views of hate crime victims, having both direct and indirect experiences had the opposite effect. These victims were relatively less empathic and more blaming of other LGBT hate crime victims. Such victim blaming is yet another indicator of the heightened traumas associated with multiple experiences of different types of hate crime victimisation. Direct and indirect experiences of hate compound the message that being LGBT is socially unacceptable and as individuals they are not worthy of respect (e.g., Noelle, 2002). With these experiences, the risk that this message becomes internalised is increased and, in turn, individuals will seek ways of avoiding future experiences of harm. One way of doing this is to enhance feelings of security and safety by dissociating oneself from certain risk factors. By placing personal blame on other victims for their experiences of anti-LGBT victimisation, some individuals can distance themselves from the perceived risk of re-victimisation – thereby protecting their emotional wellbeing in the long term (Garnets, Herek, & Levy, 1990). 26 SEM analyses generally supported the hypothesised pathways in which hate crimes impact individuals in the short term. The mediational analyses supported the proposition that direct experiences of hate crimes make people feel more vulnerable to further attacks (H2a) which, in turn, leads to greater empathy for other hate crime victims (H3a). Such findings suggest that, while direct victimisation is obviously a very negative event for individuals, it does allow them to understand and connect with fellow ingroup members. Such empathetic ties, furthermore, enhance anger and anxiety (H4a, H4B) and, consistent with IET, these predict relevant approach and avoidant behavioural responses in the short-term (H5a, H5b: e.g., anger to pro-action and anxiety to avoidance, e.g., Mackie & Smith, 2015). Media accounts of hate crimes were also shown to impact on behavioural intentions but did so via different mechanisms. Namely, instead of increasing personal vulnerability as direct experiences did, media experiences seemed to highlight that LGBT people, as a whole, live with the knowledge that they may be physically attacked (H2b: i.e., realistic threat). Perceiving the LGBT group to be under threat, individuals tended to be more empathic to victims of hate crime which, similar to direct experiences, motivate behavioural intentions via increasing feelings of anger and anxiety. However, perhaps, as the threat is less proximate than direct and indirect experiences, individuals were also less likely to engage in victim blaming, contrary to hypotheses (H3b). Feeling less ashamed was also related to less avoidance, and fewer retaliatory and security-related behaviours, as hypothesised (H5c). Feeling less anxious was related to less action altogether (partially supporting H5b). Although the link between realistic threat and blame (and its knock-on effect on behavioural intentions) was not anticipated, it suggests that while media accounts may increase behavioural intentions by promoting more ingroup empathy, the abundance of media reports of hate crimes that individuals see may also acclimatise and desensitise people to hate crimes and so blunt their intentions to act. For example, media reports may remind 27 individuals of the threat posed to their group and as such they may believe hate crimes are somewhat inevitable and unavoidable – thus they do not blame the victim, do not worry or feel ashamed over something they cannot change, and so do not feel compelled to change their situation (e.g., avoidance) or the situation of their group (e.g., pro-action). Further supporting the idea that the media may blunt individuals’ behavioural reactions, media experiences were negatively related to all actual behaviours reported three months later, serially mediated by realistic threats and victim blame. Again, media accounts of hate crime seemed to increase the saliency of the physical threat posed to LGBT group members and so the victim was not blamed as much for their victimisation and this lack of blame was associated with less action. In addition to this consistent four step mediation, media experiences were found to be indirectly associated with less pro-action (via realistic threat, blame, anxiety, and pro-action intentions) and more awareness (via realistic threat, empathy, anger, and awareness intentions). Such a finding supports the assertion that intentions are a useful predictor of behaviours (H6: Ajzen, 1991) and somewhat validates the use of behavioural intentions as a proxy for actual behaviours in previous hate crime research (e.g., Paterson et al., 2018; Walters et al., 2017). Nevertheless, intentions to avoid, retaliate, and to increase security did not predict their respective actions three months later, once all other factors had been accounted for. As individuals may find it easier to use social media and join community groups, rather than avoid certain places, change their security systems, or engage in retaliatory behaviours, these findings may reflect the differences in the ability to engage in these behaviours types, something which future research should explore. The findings have implications for practice, especially the finding that experiences of hate crimes via the media have effects on individuals and communities both in the short-term and long-term. Further research on how the media report hate crimes, and how agencies (e.g., 28 police) inform the media via press releases, is important to help mitigate the demobilising effect of media experiences revealed in this study. The results of supplementary analyses are also of theoretical and practical importance. The SEM revealed that while (direct) experiences of hate crime predict subsequent pro- active, avoidant, and awareness behaviours, behavioural intentions do not always influence later experiences (though avoidance positively predicted media experiences three months later). A pessimistic interpretation of the results could be that any reaction to minimise experiences of hate crimes is futile – and, in fact, being avoidant may be especially pointless as it was linked to more media experiences. A more optimistic view, however, is that, despite the possibility that pro-active responses, such as joining community groups, could result in greater visibility and therefore more opportunities to be victimised, we did not find this link. Instead, we believe that becoming more involved with LGBT groups and charities helps to mobilise individuals and brings communities together. With strengthened group bonds, LGBT individuals are more able to assert their rights and feel more protected, supported, and less vulnerable. Despite finding general support for our hypotheses, our research revealed some unanticipated results. Direct experiences were not associated with increases in perceptions of group threat as expected (e.g., Table 1, Figure 2) and were significantly negatively associated in the SEM. This unexpected finding may reflect a form of personalisation of their victimisation. For example, as victims are commonly attacked by someone who is known to them (Roberts, Innes, Williams, Tregidga, & Gadd, 2013), they may view themselves to be especially vulnerable to attack because of their situational circumstances, and believe that other LGBT people, with different situations, will not be the targets of such attacks. Research has also shown that hate crime offenders can be motivated by multiple factors, with many incidents occurring in the “heat of the moment”, whereby individuals caught up in an 29 altercation that is not primarily motivated by hatred towards LGBT people, ends in one person lashing out in frustration (Walters, Brown, & Wiedlitzka, 2016). In such cases, the victim may be harmed by their demonstration of hostility, but be less likely to view the incident as an attack on all LGBT people. As such, while their direct experiences lead to feelings of personal vulnerability they may be less likely to perceive it as a group-based threat. This finding was unexpected and warrants further investigation into whether there are different types of direct hate crime experiences that do and do not lead to perceptions of group-based threats. Indirect experiences did not predict any of the threat measures in the SEM as hypothesised (H2b), nor did they directly or indirectly predict any of the dependent variables, when the effects of direct and experiences were accounted for. In addition, perception of symbolic threat was not identified as a significant mediator in any of the analyses (H3a and H3b). Nevertheless, correlational analyses (Table 1) suggests that these variables were associated with relevant variables in the predicted ways and so suggest that, while both are likely to play an important role in understanding the widespread effects of hate crime, other factors may overshadow their relevance when statistically mapping the effects of hate crimes. Meanwhile anxiety was unexpectedly positively associated with feelings of blame. Similar to the blame-shame link, this association could represent a consequence of internalised homophobia in which participants not only experience reflective shame but also anxiety about who they are and the threats they and their group face (Herek et al., 1998). Anxiety was also found to be a mediator to the ‘approach’ behavioural intentions of pro- action, retaliation, and awareness raising. While this does seem to be in contrast to the existing literature which specifies anxiety to motivate avoidant tendencies (e.g., Stephan & Stephan, 1985), such links have been found before (anxiety to pro-action: Paterson et al., 2018, in press). Feeling anxious about hate crime may motivate individuals to seek support 30 from others by joining community groups (e.g., pro-action) and reaching out to people on social media (e.g., raising awareness). Furthermore, in an attempt to minimise their feelings of anxiety and powerlessness, they may want to retaliate (e.g., Van Zomeren, Fischer, & Spears, 2007). Lastly, while predictions derived from IET were generally supported cross- sectionally, they fared less well longitudinally. This could be because such responses to hate crimes are short lived. For example, if the anger felt towards the hate crime dissipates over time, its effects on behaviours would likely dissipate also. Alternatively, the complexity of our longitudinal model may have jeopardised our ability to detect these proposed mediational pathways. Future research may usefully include fewer variables over a wider temporal range (e.g., one week after, one month after, etc.) as this would better show the duration of hate crime impacts. The study was not without its limitations. Self-evidently, as a correlational design, strong causal inferences are problematic. However, this limitation was partly mitigated by the inclusion of a longitudinal element to the study, probably for the first time in research on the psychological effects of hate crime. Moreover, the findings reported here are consistent with some experimental results reported elsewhere (Paterson et al., 2018, in press) and so undue concerns about the causal direction of our findings may be misplaced. The systematic attrition from our sample might have biased the results. However, since participants who failed to complete the second questionnaire had more experiences of hate crimes, reported stronger responses and showed the same pattern of associations between the variables than did participants who completed both phases, if anything, such attrition may have led to underestimation of the effects of hate crimes. Methodologically, we only used a gay male victim within the study primes and analysed participants as ‘LGBT’ rather than their intersectional identities (e.g., gay male, 31 queer female, straight trans, etc). Although the heterogeneity of the LGBT community may give rise to different reactions based on these multifaceted identities, previous research suggests that responses to hate crimes are more likely to be derived from the LGBT superordinate identity (Paterson et al., 2018) and shared discriminatory experiences (Cortland et al., 2017) than specific sub-group identities. In addition, some of the measures were sub- optimal. Shame was measured with just one, albeit face-valid, item, while empathy was measured with four items that may be conceptually different from other definitions of empathy and, in particular, omitted any reference to perspective taking that is often central to the definition (e.g., Batson & Ahmad, 2009) Future research might be well advised to use conceptually more adequate measures. Conclusion This study supports and advances our theoretical knowledge of hate crimes whilst also having real practical implications on an issue that continues to blight individuals, communities and society as a whole. We show that hate crimes experienced directly or vicariously through other people or via the media, impact a range of perceptions, reactions and responses. The research also brings up unexpected new questions that future research could address, including the role of victim blaming and the potentially demobilising effect of media reports. Finally, as the results highlight how experiences of hate crime impact individuals, they suggest ways in which practitioners could identify, and thus manage, the resulting perceptions, emotional reactions and behavioural responses to hate crimes, whilst also helping to mobilise actions against them. 32 Footnotes 1. Participants generally reported the same gender identity (94%) and sexual orientation (87%) at both Time points. The inconsistent responses may illustrate the fluid nature of gender and sexuality. Ethnicities were also generally the same across time points (94%) and those that did differ were not greatly different (e.g., Black-British at Time 1 and Black-Other at Time 2). 2. To ensure a wide range of hate-motivated experiences were captured, participants were asked about their experiences of both hate incidents and hate crimes (those that constitute a criminal offence). For the rest of the paper, ‘hate crimes’ refer to both hate crimes and hate incidents. Also note, under UK law, verbal abuse/harassment can constitute a criminal offence (Crime and Disorder Act, 1998 (c.37)). 3. As we had not anticipated vulnerability to split into two factors, we used only one factor in the analysis. We also re-ran the analyses with all 5 items in the scale and the results showed no significant changes in associations or magnitudes. 4. Strength of LGBT identity was measured to ensure that participants were sufficiently identified as LGBT. This identity was consistent across time points (Mtime1 = 4.99 vs. Mtime2 = 5.02: t(474) = -.60, p = .547) and Identity at time 2 was not predicted by previous hate crime experiences (bs < .058, ps > .346). 5. As only 17 participants had been physically assaulted but did not know of another victim of a physical assault, we were unable to conduct the full 2x2 design. Instead, these participants were combined with participants who had direct and indirect experiences. Exclusion of these 17 participants did not alter the results. 6. As we compare the experiences of hate motivated physical assaults, we exclude media experiences from the current analyses as the item did not specify what the hate crime(s) were. 33 7. Because intended behaviours were specified as mediators to actual behaviours at Time 2, no total or direct effects were estimated for intended behaviours at Time 1. 34 References Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50(2), 179-211. Bachmann, C. L., & Gooch, B. (2017). LGBT in Britain: Hate crime and discrimination. Retrieved from http://www.stonewall.org.uk/sites/default/files/lgbt_in_britain_hate_crime.pdf Batson, C. D., & Ahmad, N. Y. (2009). Using empathy to improve intergroup attitudes and relations. Social Issues and Policy Review, 3(1), 141-177. Batson, C. D., & Ahmed, N. Y. (2009). Using empathy to improve intergroup attitudes and relations. Social Issues and Policy Review, 3, 141. Bell, J. G., & Perry, B. (2015). Outside looking in: the community impacts of anti-lesbian, gay, and bisexual hate crime. Journal of Homosexuality, 62(1), 98-120. Bibbings, L. (2004). Heterosexuality as harm: fitting in. In P. Hillyard, C. Pantazis, S. Tombs, & D. Gordon (Eds.), Beyond Criminology. Taking Harm Seriously. London: Pluto (pp. 217-235). London: Pluto. Cheng, W., Ickes, W., & Kenworthy, J. B. (2013). The phenomenon of hate crimes in the United States. Journal of Applied Social Psychology, 43(4), 761-794. College of Policing. (2014). Hate crime operational guidance. Retrieved from http://library.college.police.uk/docs/college-of-policing/Hate-Crime-Operational- Guidance.pdf Corcoran, H., & Smith, K. (2016). Hate Crime, England and Wales 2015/2016. Retrieved from https://www.gov.uk/government/statistics/hate-crime-england-and-wales-2015- to-2016 Cortland, C. I., Craig, M. A., Shapiro, J. R., Richeson, J. A., Neel, R., & Goldstein, N. J. (2017). Solidarity through shared disadvantage: highlighting shared experiences of http://www.stonewall.org.uk/sites/default/files/lgbt_in_britain_hate_crime.pdf http://library.college.police.uk/docs/college-of-policing/Hate-Crime-Operational-Guidance.pdf http://library.college.police.uk/docs/college-of-policing/Hate-Crime-Operational-Guidance.pdf https://www.gov.uk/government/statistics/hate-crime-england-and-wales-2015-to-2016 https://www.gov.uk/government/statistics/hate-crime-england-and-wales-2015-to-2016 35 discrimination improves relations between stigmatized groups. Journal of Personality and Social Psychology, 113(4), 547-567. Cottrell, C. A., & Neuberg, S. L. (2005). Different emotional reactions to different groups: a sociofunctional threat-based approach to "prejudice". Journal of Personality and Social Psychology, 88, 770-789. Crime and Disorder Act. (1998 (c.37)). D’Augelli, A. R., Grossman, A. H., & Starks, M. T. (2006). Childhood gender atypicality, victimization, and PTSD among lesbian, gay, and bisexual youth. Journal of Interpersonal Violence, 21(11), 1462-1482. Garnets, L., Herek, G. M., & Levy, B. (1990). Violence and victimization of lesbians and gay men: Mental health consequences. Journal of Interpersonal Violence, 5(3), 366-383. Herek, G. M., Cogan, J. C., & Gillis, J. R. (2002). Victim experiences of hate crimes based on sexual orientation. Journal of Social Issues, 58, 319-339. Herek, G. M., Cogan, J. C., Gillis, J. R., & Glunt, E. K. (1998). Correlates of internalized homophobia in a community sample of lesbians and gay men. Journal of the Gay and Lesbian Medical Association, 2, 17-26. Herek, G. M., Gillis, J. R., & Cogan, J. C. (1999). Psychological sequelae of hate-crime victimization among Lesbian, Gay and Bisexual adults. Journal of Consulting and Clinical Psychology, 67, 945-951. Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6(1), 1-55. Iganski, P., & Lagou, S. (2015). Hate crimes hurt some more than others: Implications for the just sentencing of offenders. Journal of Interpersonal Violence, 30(10), 1696-1718. 36 Iyer, A., & Leach, C. W. (2008). Emotion in inter-group relations. European Review of Social Psychology, 19, 86-125. Iyer, A., Schmader, T., & Lickel, B. (2007). Why individuals protest the perceived transgressions of their country: the role of anger, shame and guilt. Personality and Social Psychology Bulletin, 33, 572-587. Kelloway, E. K. (2014). Using Mplus for structural equation modeling: A researcher's guide: Sage Publications. Leonard, D. J., Moons, W. G., Mackie, D. M., & Smith, E. R. (2011). “We’re mad as hell and we’re not going to take it anymore”: Anger self-stereotyping and collective action. Group Processes & Intergroup Relations, 14(1), 99-111. Mackie, D. M., Devos, T., & Smith, E. R. (2000). Intergroup emotions: explaining offensive action tendencies in an intergroup context. Journal of Personality and Social Psychology, 79(4), 602-616. Mackie, D. M., & Smith, E. R. (Eds.). (2015). Intergroup emotions. Washington DC: American Psychological Association. Mackie, D. M., Smith, E. R., & Ray, D. G. (2008). Intergroup emotions and intergroup relations. Social and Personality Psychology Compass, 2(5), 1866-1880. doi:10.1111/j.1751-9004.2008.00130.x MacKinnon, D. P., Krull, J. L., & Lockwood, C. M. (2000). Equivalence of the mediation, confounding and suppression effect. Prevention science, 1(4), 173-181. Muthén, L. K., & Muthén, B. O. (2011). Mplus User's Guide. Sixth Edition. Los Angeles: Muthén & Muthén. Noelle, M. (2002). The ripple effect of the Matthew Shephard murder: impact on the assumptive worlds of members of the targeted group. American Behavioral Scientist, 46, 27-50. 37 Paterson, J. L., Brown, R., & Walters, M. A. (2018). Understanding victim group responses to hate crime: shared identities, perceived similarity and intergroup emotions. Testing, Psychometrics, Methodology in Applied Psychology, 25(2), 163-177. doi:10.4473/TPM25.2.1 Paterson, J. L., Brown, R., & Walters, M. A. (in press). Feeling for and as a group member: understanding LGBT victimisation via group-based empathy and intergroup emotions. British Journal of Social Psychology. doi:10.1111/bjso.12269 Perry, B. (2009). Hate Crimes: The Consequences of Hate Crime: Praeger Publishers. Perry, B., & Alvi, S. (2012). 'We are all vulnerable’: The 'in terrorem' effects of hate crimes. International Review of Victimology, 18, 57-71. Roberts, C., Innes, M., Williams, M. L., Tregidga, J., & Gadd, D. (2013). Understanding who commits hate crime and why they do it. Retrieved from https://orca.cf.ac.uk/58880/1/understanding-who-commits-hate-crime-and-why-they- do-it-en.pdf Savin-Williams, R. C., & Ream, G. L. (2007). Prevalence and stability of sexual orientation components during adolescence and young adulthood. Archives of Sexual Behavior, 36(3), 385-394. Sheikh, S. (2014). Cultural variations in shame’s responses: A dynamic perspective. Personality and Social Psychology Review, 18(4), 387-403. Smith, E. R. (1993). Social identity and social emotions: toward new conceptualizations of prejudice. In D. M. Mackie & D. L. Hamilton (Eds.), Affect, cognition and stereotyping (pp. 297-315). San Diego: Academic Press. Stephan, W. G. (2014). Intergroup anxiety theory, research, and practice. Personality and Social Psychology Review, 18(3), 239-255. https://orca.cf.ac.uk/58880/1/understanding-who-commits-hate-crime-and-why-they-do-it-en.pdf https://orca.cf.ac.uk/58880/1/understanding-who-commits-hate-crime-and-why-they-do-it-en.pdf 38 Stephan, W. G., & Stephan, C. W. (1985). Intergroup anxiety. Journal of Social Issues, 41, 157-175. Tajfel, H., & Turner, J. (1979). An Integrative Theory of Intergroup Conflict. In W. G. Austin & S. Worchel (Eds.), The Social Psycholgy of Intergroup Relations (pp. 33-47). California: Brooks & Cole. Tangney, J. P. (1990). Assessing individual differences in proneness to shame and guilt: Development of the Self-Conscious Affect and Attribution Inventory. Journal of Personality and Social Psychology, 59(1), 102-111. Tangney, J. P., Wagner, P., Fletcher, C., & Gramzow, R. (1992). Shamed into anger? The relation of shame and guilt to anger and self-reported aggression. Journal of Personality and Social Psychology, 62, 669-675. Van Zomeren, M., Fischer, A. H., & Spears, R. (2007). Testing the limits of tolerance: How intergroup anxiety amplifies negative and offensive responses to out-group-initiated contact. Personality and Social Psychology Bulletin, 33(12), 1686-1699. Walters, M. A., Brown, R., & Wiedlitzka, S. (2016). Causes and motivations of hate crime. Retrieved from https://www.equalityhumanrights.com/sites/default/files/research- report-102-causes-and-motivations-of-hate-crime.pdf Walters, M. A., Paterson, J. L., Brown, R., & McDonnell, L. (2017). Hate crimes against trans people: assessing emotions, behaviors and attitudes towards criminal justice agencies. Journal of Interpersonal Violence. doi:10.1177/0886260517715026 https://www.equalityhumanrights.com/sites/default/files/research-report-102-causes-and-motivations-of-hate-crime.pdf https://www.equalityhumanrights.com/sites/default/files/research-report-102-causes-and-motivations-of-hate-crime.pdf 39 Figure 1. Hypothesised model (H2 -H6) Retaliation Security Pro-action Avoidance Anxiety Anger Empathy Blame Indirect Direct Media Awareness Real Threat Vuln Actual ret’n Actual pro-act Actual avoidance Actual awaren’ T2 Measured at T1 Shame Actual Security Previous experiences Perceptions of threat Perceptions of victim Emotions to crime Behavioural intentions Actual behaviours - Symb Threat 40 Figure 2. MANCOVA comparing the effects of direct, indirect and no experiences of hate crimes at Time 1 1 2 3 4 5 6 7 A gr ee m en t No experience (n = 391) Indirect only (n = 219) Direct and indirect (n = 163) 41 Table 1 Means, SDs, and correlations of variables used in SEM 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 1. Direct - .64** .20** .34** -.01 .07 -.14** .41** .04 .32** .23** .45** .52** .35** .33** .24** .28** .25** .24** .32** .19** 2. Indirect - .30** .33** .07 .13** .01 .20** .12* .31** .11* .34** .45** .40** .39** .16** .17** .16** .20** .28** .10* 3. Media - .22** .28** .21** .16** -.10* .17** .17** -.02 .06 .09 .23** .23** -.08 .01 .02 .04 .15** .01 4. Vuln - .39** .37** .26** .08 .26** .65** .07 .53** .42** .39** .39** .22** .11* .15** .13** .17** -.01 5. Real - .73** .46** -.28** .45** .23** -.13* .11* .08 .26** .31** -.10* -.12* -.06 .05 .06 -.12* 6. Symb - .43** -.16** .45** .36** -.04 .22** .19** .34** .34** -.01 -.01 .01 .10* .09 -.09 7. Empathy - -.37** .61** .23** -.08 .05 .01 .20** .31** -.12* -.05 -.10* .01 .02 -.15** 8. Blame - -.23** .16** .41** .37** .36** .16** .08 .36** .37** .30** .21** .19** .26** 9. Anger - .31** -.01 .12** .14** .32** .39** -.04 .03 -.02 .09* .15** -.05 10. Anxiety - .23** .58** .52** .43** .36** .20** .18** .17** .18** .21** .03 11. Shame - .29** .27** .16** .04 .19** .19** .16** .10* .11* .13* 12. Avoid B - .73** .45** .38** .33** .23** .19** .21** .21** .08 13. Sec BI - .57** .47** .33** .24** .26** .22** .23** .12** 14. Pro-act BI - .76** .28** .15** .17** .24** .28** .05 15.Aware BI - .26** .16** .10* .22** .32** .01 16. Retal BI - .19** .18** .20** .20** .21** 17.Actual Avoid - .454** .43** .33** .32** 42 18 Actual Security - .54** .42** .59** 19.Actual Pro-act - .67** .59** 20.Actual aware - .41** 21.Actual retal - Means (SDs) 1.30 (.36) 1.50 (.42) 2.55 (1.11) 3.72 (1.67) 5.32 (1.34) 5.04 (1.54) 5.94 (1.12) 1.84 (1.47) 5.56 (1.29) 3.70 (1.69) 2.89(1.98) 2.56 (1.63) 2.76 (1.73) 3.52 (1.73) 3.79 (1.80) 1.36 (.48) 4.08 (.73) 3.71 (1.17) 3.94 (.91) 4.10 (.98) 3.69 (1.27) *p < .01,**p < .001 Note. Variables 1-16 measured at Time 1, variables 17-21 measured at T2. 43 Figure 3. Structural equation model *p < .05, **p < .01, ***p < .001. Retaliation Security Pro-action Avoidance Anxiety Anger Empathy Blame Indirect Direct Media Awareness Real Vuln Symb Actual ret’n Actual pro-act Actual avoidance Actual awaren’ .40** -.30* .11* .28*** .17** .25*** .33*** -.57*** .20** .17* .16*** .35*** .46** .49*** .72*** .38*** .24*** .29*** .33*** .61*** .55*** .18*** .16*** .14** .09* .17*** .22*** T2 Measured at T1 -.14** Shame Actual Security .21*** -.12* .22* .44*** .18* .18* .15* Previous experiences Perceptions of threat Perceptions of victim Emotions to crime Behavioural intentions Actual behaviours 44 Not shown: Age was used as a covariate. All factors at Time 1 estimated actual behaviours at Time 2. All items loaded onto their respective factors (p < .001), factors at the same level were allowed to covary, two items of realistic threats, empathy, and actual avoidance scales were covaried. χ2(1338) = 2684.757, p < .001 (MLR scaling factor = 1.114), CFI = .921, RMSEA = .047 [95% CIs: .044/.049], SRMR = .087. 45 Table 2. Indirect effects of hate crime experiences on behavioural intentions at Time 1 Predictor Outcome Total indirect effect [95% CIs] Significant indirect effect pathways [95% CIs] Media experiences Avoidance BI .001 [-.016/.017] Media  realistic threat  empathy  anger  avoid BI: -.009 [-.017/-.001] Media  realistic threat  empathy  anxiety  avoid BI: .027 [.009/.044] Media  realistic threat  blame  shame  avoid BI: -.013 [-.022/-.004] Media  realistic threat  blame  anxiety  avoid BI: -.033 [-.049/-.016] Security BI .011 [-.007/.030] Media  realistic threat empathy  anxiety  security BI: .024 [.009/.039] Media  realistic threat  blame  shame  security BI: -.012 [-.020/-.003] Media  realistic threat  blame  anxiety  security BI: -.030 [-.044/-.015] Pro-action BI .033 [.013/.053] Media  realistic threat  empathy  anger  pro-action BI: .016 [.006/.026] Media  realistic threat  empathy  anxiety  pro-action BI: .014 [.004/.025] Media  realistic threat  blame  anxiety  pro-action BI: -.018 [-.028/-.007] Awareness BI .049 [.025/.072] Media  realistic threat  empathy  anger  awareness BI: .025 [.011/.038] Media  realistic threat  empathy  anxiety  awareness BI: .013 [.002/.023] 46 Media  realistic threat  blame  anxiety  awareness BI: -.015 [-.026/-.005] Retaliation BI -.010 [-.022/.003] Media  realistic threat  blame  shame  retaliate BI: -.010 [-.018/-.002] Media  realistic threat  blame  anxiety  retaliate BI: -.011 [-.020/-.003] Direct experiences Avoidance BI .060 [.013/.106] Direct  vulnerability  empathy  anger  avoid BI: -.006 [-.012/-.001] Direct  vulnerability  empathy  anxiety  avoid BI: .019 [.003/.035] Security BI .049 [.003/.095] Direct  vulnerability  empathy  anxiety  security BI: .017 [.003/.031] Pro-action BI .016 [-.032/.064] Direct  vulnerability  empathy  anger  pro-action BI: .011 [.001/.021] Direct  vulnerability  empathy  anxiety  pro-action BI: .010 [.001/.019] Awareness BI -.002 [-.065/.061] Direct  vulnerability  empathy  anger  awareness BI: .017 [.004/.031] Retaliation BI .032 [.003/.060] None Indirect experiences Avoidance BI .006 [-.028/.039] None Security BI .014 [-.022/.050] None Pro-action BI .029 [-.015/.074] None Awareness BI .041 [-.020/.101] None Retaliation BI -.005 [-.027/.017] None Note. BI = Behavioural intention. CIs that do not straddle zero are significant. 47 Table 3. Direct, indirect and total effects of the longitudinal SEM Predictor Outcome Total estimate [95% CIs] Direct effect [95% CIs] Total indirect effect [95% CIs] Significant indirect effect pathways [95% CIs] Media experiences Avoidance .054 [-.069/.177] .120 [.000/.240] -.066 [-.117/-.015] Media  realistic threat blame actual avoidance: -.069 [-.118/-.021] Security -.049 [-.390/.042] -.019 [-.114/.076] -.030 [-.067/.007] Media realistic threat  blame  actual security: -.033 [-.056/-.011] Pro-action -.032 [-.105/.040] -.041 [-.100/.018] .009 [-.022/.041] Media  realistic threat  blame  actual pro-action: -.028 [-.051/-.004] Media  realistic threat  blame anxiety  pro-action intention  actual pro-action: -.003 [-.005/.-.001] Awareness .061 [-.036/.158] .057 [-.038/.152] .004 [-.030/.038] Media  realistic threat  blame  actual awareness: -.023 [-.044/-.002] Media  realistic threat empathy  anger awareness intention  actual awareness: .005 [.001/.010] 48 Retaliation -.001 [-.074/.073] .033 [-.038/.105] -.034 [-.066/-.002] Media  realistic threat  blame  actual retaliation: -.024 [-.050/-.005] Direct experiences Avoidance .576 [.117/1.034] .420 [-.048/.887] .156 [-.007/.319] None Security .209 [-.046/.464] .084 [-.207/.376] .125 [.010/.240] None Pro-action .170 [-.066/.407] .184 [-.089/.458] -.014 [-.119/.091] None Awareness .245 [.008/.481] .243 [-.036/.521] .002 [-.108/.112] None Retaliation .193 [-.014/.401] .163 [-.075/.401] .030 [-.070/.131] None Indirect experiences Avoidance -.269 [-.686/.147] -.230 [-.633/.173] -.039 [-.133/.055] None Security -.092 [-.329/.146] -.074 [-.312/.165] -.018 [-.090/.054] None Pro-action .011 [-.199/.221] -.001 [-.226/.225] .012 [-.034/.057] None Awareness -.030 [-.305/.246] -.032 [-.322/.258] .002 [-.045/.049] None Retaliation -.122 [-.329/.086] -.096 [-.305/.114] -.026 [-.079/.027] None Note. CIs that do not straddle zero are significant. 49 Figure 4 Cross-lagged SEM of experiences and behavioural intentions, age as a covariate Note. Start of arrow indicates Factor measured at Time 1, point of arrow indicates Factor measured at Time 2. †p = .06, *p < .05. Model fit: χ2(506) = 982.532, p < .001 (MLR scaling factor = 1.13), CFI = .955, RMSEA = .045 [95% CIs: .041/.049], SRMR = .043. Retaliation Security Pro-action Avoidance Media Indirec Direct Awareness .24† .21† .58* .23*