INTRODUCTION
Since the Stonewall Uprising, there have been numerous instances of anti-LGBTQ+ violence in the United States. Despite progress on LGBTQ+ rights (Flores Reference Flores2014), anti-LGBTQ+ violence and hate crimes have increased,Footnote 1 while several states introduced a record number of anti-LGBT+ laws recently.Footnote 2 Perhaps the most prominent, recent, instance of anti-LGBTQ+ violence was the 2022 Club Q massacre, where a gunman killed five clubgoers at a Colorado Springs LGBTQ+ nightclub. These violent acts, while sympathy-inducing within media and among some political elites, may reflect durable heteronormative societal norms.Footnote 3 Therefore, an open question is whether indirect (i.e., media observation of violence) exposure to high-profile civilian-perpetrated violence against LGBTQ+ group members motivates introspection among the mass public, shifting attitudes prosocially toward LGBTQ+ people.
We synthesize several theoretical insights and present a Fickle Prosocial Violence Response (FPVR) model to explain how violence against marginalized groups may elicit prosocial attitudes toward targeted groups. Although the mass public may not strongly empathize with marginalized minority groups (Cikara et al. Reference Cikara, Bruneau, Van Bavel and Saxe2014), violence against marginalized groups may elicit prosocial attitudes if the violence is salient, perceptibly illegitimate, and the media and/or elites respond sympathetically (Birkland Reference Birkland1998; Branscombe and Miron Reference Branscombe, Miron, Tiedens and Leach2004; Harth, Kessler, and Leach Reference Harth, Kessler and Leach2008; Iyengar Reference Iyengar1994; Vossen, Piotrowski, and Valkenburg Reference Vossen, Piotrowski and Valkenburg2017). However, prosocial attitude adoption may be short-term. Social group attitudes are typically entrenched, even in light of salient events (Kite, Togans, and Schultz Reference Kite, Togans, Schultz and Keith2019; Sears Reference Sears, Shanto and William1993; Tuch and Weitzer Reference Tuch and Weitzer1997). Immediate adoption of prosocial beliefs after violence may be counterbalanced by countervailing information in a discriminatory society (Vuletich and Keith Payne Reference Vuletich and Keith Payne2019). Elite messaging and pressure to support targeted groups may dissipate after an event loses salience (Downs Reference Downs1972), undercutting sustainable prosocial attitudinal shifts (Birkland and Lawrence Reference Birkland and Lawrence2009; Zaller Reference Zaller1992).
We find evidence supporting the FPVR model by using several surveys and an unexpected-event-during-survey-design (UESD). Studies 1–3 demonstrate the public adopts prosocial attitudes toward segments of the LGBTQ+ community and their political rights shortly after civilian violence against LGBTQ+ group members—that is, the Pulse massacre and Matthew Shepard’s murder. However, these attitudinal shifts do not persist. Study 4 demonstrates the Club Q massacre had no effect on anti-gay or anti-trans attitudes. Consistent with the model, we provide evidence that the null effects at the outset are due to the less salient nature of the Club Q massacre vis-à-vis the Pulse massacre and Shepard’s murder. We provide corroborating evidence by demonstrating that less salient violent incidents against LGBTQ+ people outside those in Studies 1–4 largely do not motivate prosocial mass attitudes.
Our theory and evidence make several contributions. First, the FPVR model helps explain how violence against marginalized groups motivates prosocial beliefs toward targeted groups among the mass public. Our findings are important in light of several salient instances of civilian violence against marginalized groups in the United States: Vincent Chin’s 1982 murder, a Chinese man killed on the basis of anti-Japanese resentment; James Byrd’s 1996 murder, a Texas Black man lynched by white supremacists; the 2015 Charleston Church massacre, where a white supremacist murdered nine Black churchgoers; the 2015 Stanford sexual assault case (People v. Turner), where a Stanford undergraduate man sexually assaulted a woman; the 2018 Pittsburgh Tree of Life synagogue shooting, where a man killed 11 Jews on the basis of the anti-semitic “white genocide” conspiracy theory; the 2019 El Paso massacre, where a white supremacist killed 23 people, mostly Latinos, to counteract a “Hispanic invasion”; the 2021 Atlanta spa shooting, where eight people, mostly Asian women, were killed; the 2022 Buffalo massacre, where a white supremacist killed 10 Black people because he felt non-whites were “replacing” whites; and the 2023 shooting of Hisham Awartani, Tahseen Aliahmad, and Kinnan Abdalhamid, three Palestinian young men who were shot by a white man while speaking Arabic and wearing keffiyehs. We show that these events may not serve as sustainable moments of reevaluation concerning the sociopolitical status of marginalized groups and may not motivate prosocial attitudes at the outset if they are insufficiently salient. Thus, our model and evidence may explain why these events have not led to societal adjustment of beliefs perpetuating social inequalities and intergroup hostility.
Second, our analysis extends prior research on violence against marginalized groups by examining a different perpetrator type (civilian) and group (LGBTQ+). Prior research on violence and prosocial attitudes in the United States typically focuses on state (i.e., police) violence against Black people. This research often identifies prosocial responses to violence but mixed evidence on effect sustainability (Chudy and Jefferson Reference Chudy and Jefferson2021; Reny and Newman Reference Reny and Newman2021; Sigelman et al. Reference Sigelman, Welch, Blesdoe and Combs1997; Tuch and Weitzer Reference Tuch and Weitzer1997). Civilian-perpetrated violence against LGBTQ+ group members may have theoretically distinct but important consequences. Civilian-perpetrated (instead of state-perpetrated) violence may be less likely to initially and/or sustainably motivate prosocial attitudes. The mass public may attribute state violence to systemic yet reformable institutional problems, motivating policy preferences benefiting targeted groups (Oskooii Reference Oskooii2016). Yet civilian violence may be rationalized as a problem inherent to a troubled individual as opposed to the public’s systemic aggregate queerphobia (Ott and Aoki Reference Ott and Aoki2002), undercutting, at worst, initial introspection over one’s own queerphobic beliefs post-violence, at best, sustained introspection in a heteronormative society consistently encouraging queerphobia.Footnote 4 Moreover, unlike racialized state violence, the violence we examine are not associated with subsequent mass protest, which may sustain event salience and facilitate long-lasting attitudinal shifts (Reny and Newman Reference Reny and Newman2021). Consistent with these theoretical perspectives (and the FPVR model), our evidence highlights similarities and contrasts in the prosocial consequences of different types of violence, paving the way for further work in assessing how contextual variation of violent events may differentially motivate mass attitudinal responses.
Third, our analysis contributes to the Focusing Event literature (Birkland Reference Birkland1998). Prior research shows salient events shift mass attitudes, but briefly because of eventual salience loss (Birkland and Lawrence Reference Birkland and Lawrence2009; Sigelman et al. Reference Sigelman, Welch, Blesdoe and Combs1997). Additionally, LGBTQ+ politics research demonstrates high-profile pro-LGBTQ+ court cases (Flores Reference Flores2015), Pride parades (Ayoub, Page, and Whitt Reference Ayoub, Page and Whitt2021), and celebrities coming out (Miller et al. Reference Miller, Flores, Haider-Markel, Lewis, Tadlock and Taylor2020), can motivate prosocial attitudes toward LGBTQ+ people. But this research places little emphasis on effect sustainability, does not assess event salience variation at the outset, and does not focus on violence against LGBTQ+ people, which may reflect, instead of undercut, queerphobia. We provide new evidence consistent with Focusing Event Theory in an unexplored domain.
VIOLENCE AND PROSOCIALITY
Preexisting theory and evidence demonstrates direct or proximal (i.e., via close social ties, like family, friends, and acquaintances) violence exposure during intergroup conflict may motivate parochialism, encourage intra- but not intergroup altruism, and undercut emotional substrates facilitating intergroup prosocial behaviors and attitudes, including, positive evaluations of out-groups and support for their political rights (Hadzic, Carlson, and Tavits Reference Hadzic, Carlson and Tavits2020; Lupu and Peisakhin Reference Lupu and Peisakhin2017; Mironova and Whitt Reference Mironova and Whitt2018; Rusch Reference Rusch2014). Other evidence, building on Post-Traumatic Growth and Altruism Born of Suffering Theory (Staub and Vollhardt Reference Staub and Vollhardt2010), shows intergroup violence can motivate prosocial, altruistic attitudes and behaviors toward out-groups (Bakke, O’Loughlin, and Ward Reference Bakke, O’Loughlin and Ward2009). Direct or proximal violence exposure may motivate intergroup prosociality since victimization generates a basis for empathy (Sirin, Valentino, and Villalobos Reference Sirin, Valentino and Villalobos2021).
Although prior work suggests direct or proximal exposure to intergroup, mostly interethnic, violence motivates prosociality, it is less clear how one-sidedFootnote 5 indirect exposure to violence against LGBTQ+ people influences prosocial attitudes toward LGBTQ+ group members among dominant groups or the mass public. Hereafter, we define prosocial attitudes as positive feelings toward LGBTQ+ group members and policies facilitating their rights.
One expectation is that indirect exposure to one-sided violence may not motivate prosocial beliefs. Insufficient media coverage and attention to violent events may not produce agenda-setting effects mobilizing prosocial mass attitudes (Birkland Reference Birkland1998). Additionally, Social Identity Theory (SIT) implies dominant group members garner self-esteem from minority group marginalization (Tajfel and Turner Reference Tajfel and Turner1982). Thus, the mass public may garner psychic benefits from indirectly observing violence against minority groups (Cikara et al. Reference Cikara, Bruneau, Van Bavel and Saxe2014). Consistent with Intergroup Emotions Theory (IET), these dynamics may be exacerbated by the absence of direct experiences with analogous violence facilitating empathy (Sirin, Valentino, and Villalobos Reference Sirin, Valentino and Villalobos2021). Moreover, the social distance between modal mass public members and, for example, LGBTQ+ people may generate an empathy gap,Footnote 6 undercutting the adoption of prosocial attitudes after indirect violence exposure (Cikara et al. Reference Cikara, Bruneau, Van Bavel and Saxe2014). Finally, if the violence is civilian-perpetrated, the violent event may be framed by the media as a problem inherent to a troubled individual instead of societal antipathy toward LGBTQ+ people (Iyengar Reference Iyengar1994; Ott and Aoki Reference Ott and Aoki2002; Zahzah Reference Zahzah2019), which could undercut reflection concerning one’s own antipathic beliefs among the mass public. Therefore, we may observe an empirical pattern consistent with Figure 1a, where indirect exposure to civilian violence against marginalized groups does not motivate mass prosocial attitudes toward targeted groups.
Figure 1. Stylized Expectations Concerning the Effect of Violence against Marginalized Groups on Prosocial Attitudes
Note: Horizontal lines denote prosocial attitudes toward marginalized groups (y-axis) over time (x-axis). The dotted line characterizes violence against a marginalized group.
Another expectation is that, under some conditions, indirect exposure to violence against marginalized groups may motivate prosocial attitudes to ameliorate conditions concomitant with the violence. Focusing Event Theory implies salient violent incidents can mobilize mass attitudes (Birkland Reference Birkland1998). These attitudes may be more likely to be mobilized prosocially if the media and elites express the violence is illegitimate and are sympathetic toward the targeted group (Iyengar Reference Iyengar1994; Zaller Reference Zaller1992). Indeed, sympathetic messaging by partisan elites post-violence may help socially conservative co-partisans reconsider prejudicial attitudes (Harrison and Michelson Reference Harrison and Michelson2017). The media also has a powerful influence on LGBTQ+ mass attitudes. Positive LGBTQ+ media portrayals and parasocial LGBTQ+ contact motivates prosocial attitudes toward LGBTQ+ people (Ayoub and Garretson Reference Ayoub and Garretson2017; Miller et al. Reference Miller, Flores, Haider-Markel, Lewis, Tadlock and Taylor2020).
Likewise, alternative SIT and IET insights suggest if the mass public feels one-sided civilian violence against marginalized groups is illegitimate, it reflects poorly on their own stigmatizing beliefs, even if minority group marginalization otherwise facilitates self-esteem (Harth, Kessler, and Leach Reference Harth, Kessler and Leach2008). Dominant group or mass public members may emotionally regulate these psychic costs by reacting to violence against marginalized groups with sympathy and/or empathy (Branscombe and Miron Reference Branscombe, Miron, Tiedens and Leach2004), motivating the downstream adoption of prosocial attitudes toward marginalized groups (Harth, Kessler, and Leach Reference Harth, Kessler and Leach2008; Stotzer Reference Stotzer2009).
Some prior research implies prosocial attitude adoption toward marginalized groups after violence exposure may be durable. The mass public has become increasingly inclusive toward segments of the LGBTQ+ community over several decades (Flores Reference Flores2014), suggesting the public may be durably receptive to sympathetic appeals after violence against LGBTQ+ group members. Indeed, Broockman and Kalla (Reference Broockman and Kalla2016) show a perspective-taking exercise can increase support for transgender antidiscrimination policies up to 3 months. Oskooii, Lajevardi, and Collingwood (Reference Oskooii, Lajevardi and Collingwood2021) show high-profile institutionalized discrimination against religious minorities can reduce mass support for policies negatively affecting targeted groups up to a year. Reny and Newman (Reference Reny and Newman2021) show anti-Black police violence can motivate prosocial attitudes toward Black people up to at least 100 days. Therefore, we might observe an empirical pattern consistent with Figure 1b, where the public adopts increasingly prosocial attitudes after indirect exposure to civilian violence against marginalized groups, and these attitudinal shifts are durable.
THE FICKLE PROSOCIAL VIOLENCE RESPONSE MODEL
We develop and present an FPVR model, which posits perceptibly illegitimate salient civilian violence against marginalized groups can motivate prosocial attitudes toward targeted groups. But these attitudinal shifts may be fickle given reductions in event salience, the dispositional qualities of social group attitudes, and countervailing information in an otherwise discriminatory society.
Issue-Attention Cycle Theory posits that the public may react to dramatic events highlighting ignored social issues, like violence against LGBTQ+ group members, in an initially proactive manner. However, attitudinal shifts seeking to resolve a social ill may not be sustainable when it becomes clear resolving the problem is difficult (e.g., reevaluating queerphobic beliefs offering a privileged status) and the problem becomes less salient over time (Downs Reference Downs1972). Prior research implies sympathetic media and elite messaging after violence must persist to generate sustainable prosocial responses (Zaller Reference Zaller1992). Without persistent sympathetic messaging, the masses may not be continually encouraged to adopt positive emotions toward targeted groups in addition to an understanding of the violence as illegitimate, producing a decay in prosocial attitudinal responses.
Moreover, prosocial attitudinal responses may be short-term impression management. Illegitimate violence rejected by society, media, and elites may motivate prosocial expressions toward the targeted group among the masses to save face (Harth, Kessler, and Leach Reference Harth, Kessler and Leach2008), but may not result in long-term attitudinal shifts motivated by the difficult task of dismantling hierarchical social relations (Nguyen et al. Reference Nguyen, Criss, Michaels, Cross, Michaels, Dwivedi and Huang2021). Short-term impression management may not be capable of undermining predispositions toward marginalized groups rooted in preadult socialization (Kite, Togans, and Schultz Reference Kite, Togans, Schultz and Keith2019; Sears Reference Sears, Shanto and William1993). Long-term attitudinal shifts may also be undercut by countervailing pressure to adhere to queerphobic norms in an otherwise heteronormative society (Vuletich and Keith Payne Reference Vuletich and Keith Payne2019).
Framing theory may also help explain the potential absence of long-term prosocial attitudinal shifts. Story framing affects how the public assigns responsibility to an event and preferred policy and societal responses. Media outlets may adopt episodic or thematic frames in their news coverage. Episodic frames emphasize event-centered information with attention toward an individual’s actions (e.g., the violent perpetrator), whereas thematic frames emphasize broader problems (e.g., queerphobia) (Iyengar Reference Iyengar1994). Ott and Aoki (Reference Ott and Aoki2002) and Zahzah (Reference Zahzah2019) posit media frames of prominent instances of violence against LGBTQ+ people, such as Matthew Shepard’s murder and the Pulse massacre, often emphasize the perpetrator’s gratuitous violence instead of societal heteronormativity. These episodic frames may allow mass public members to simply express prosocial attitudes toward LGBTQ+ to absolve oneself of short-term guilt but lose sight of reflecting over their quotidian role facilitating a heteronormative society in the long term (Ott and Aoki Reference Ott and Aoki2002), especially in light of countervailing information from a queerphobic society.Footnote 7
In summary, an observable implication of the theoretical synthesis informing the FPVR model is that indirect exposure to salient and sympathetic messaging from media and elites after violence against LGBTQ+ group members may encourage the adoption of prosocial attitudes toward segments of the LGBTQ+ community. But, the adoption of prosocial attitudes toward LGBTQ+ group members may not be long-lasting. Therefore, we may observe an empirical pattern consistent with the solid line in Figure 1c. H1: Indirect exposure to civilian violence against LGBTQ+ group members will initially increase prosocial attitudes toward LGBTQ+ group members. H2: But indirect exposure to civilian violence against segments of the LGBTQ+ community will not produce sustainable increases in prosocial attitudes.
Prior evidence corroborates the FPVR model. Some evidence shows high-profile anti-Black police violence increased prosocial attitudes toward Black people, but these attitudes reverted to the pre-violence equilibrium shortly thereafter (Chudy and Jefferson Reference Chudy and Jefferson2021; Nguyen et al. Reference Nguyen, Criss, Michaels, Cross, Michaels, Dwivedi and Huang2021; Tuch and Weitzer Reference Tuch and Weitzer1997). Birkland and Lawrence (Reference Birkland and Lawrence2009) demonstrate Columbine immediately increased gun control support, but only briefly.
Individual-Level Heterogeneity
Shared Marginalization
Group Empathy Theory posits marginalized group members who possess similar discriminatory experiences support each other (Sirin, Valentino, and Villalobos Reference Sirin, Valentino and Villalobos2021). Cross-group support may be more likely if the discrimination a particular group experiences is perceptibly shared (Cortland et al. Reference Cortland, Craig, Shapiro, Richeson, Neel and Goldstein2017). Members of other subjugated groups (e.g., women) may perceive similarities between their experiences and those of LGBTQ+ group members, especially with regard to targeted violence. Indeed, we have argued in the Introduction that women and non-whites have been historically subject to targeted violence in a conceivably similar manner as LGBTQ+ people. Thus, group members discriminated against on other dimensions, like race and/or gender, may be more inclined to respond prosocially toward LGBTQ+ group members after exposure to violence against segments of the LGBTQ+ community.
Political Liberalism
Relative to conservatives and moderates, liberals are less socially conservative concerning sexuality and gender and are more accepting of marginalized social groups. Indeed, liberals are more favorable toward segments of the LGBTQ+ community and pro-LGBTQ+ policies (Flores Reference Flores2014). Conservatives are more likely to adopt anti-LGBTQ+ beliefs in response to threatening anti-LGBTQ+ elite rhetoric, while liberals are resistant to such rhetoric (Górska and Tausch Reference Górska and Tausch2022). Relative to moderates and conservatives, liberals are also more inclined to respond prosocially toward marginalized groups in response to high-profile state violence against said groups (Reny and Newman Reference Reny and Newman2021). Therefore, liberals may be more likely than conservatives to adopt prosocial attitudes toward LGBTQ+ group members in response to violence against LGBTQ+ people.
Geographic Context
Individuals living in areas with a higher concentration of LGBTQ+ people may be more likely to come into contact with LGBTQ+ group members and to therefore develop strong social ties with LGBTQ+ people (Tadlock et al. Reference Tadlock, Flores, Haider-Markel, Lewis, Miller and Taylor2017). Harrison and Michelson (Reference Harrison and Michelson2019) identify consistent evidence that contact with LGBTQ+ group members motivates prosociality toward different segments of the LGBTQ+ community. Given that individuals living in areas with more LGBTQ+ people may be dispositionally favorable toward the LGBTQ+ community (Thompson Reference Thompson2022), they may also be more inclined to adopt prosocial attitudes toward segments of the LGBTQ+ community after high-profile civilian violence against LGBTQ+ group members. Indeed, prior research shows that individuals living in LGBTQ+ geographic contexts resist anti-LGBTQ+ elite rhetoric (Górska and Tausch Reference Górska and Tausch2022).
In summary, H3a–c: indirect exposure to civilian violence against LGBTQ+ group members will be more likely to motivate prosocial attitudes toward LGBTQ+ group members among: (a) non-whites and women relative to whites and men; (b) liberals relative to moderates and conservatives; and (c) individuals living in geographic contexts with more LGBTQ+ people relative to those living in contexts with less LGBTQ+ people.
Event-Level Salience Heterogeneity
The FPVR model implies violent events must be sufficiently salient (i.e., covered by media and paid attention to by the public) to generate attitudinal shifts toward targeted groups (Birkland Reference Birkland1998; Downs Reference Downs1972; Zaller Reference Zaller1992). Indeed, prior studies demonstrating mass attitudinal shifts after U.S. violent events are analyzing high-profile events (Birkland and Lawrence Reference Birkland and Lawrence2009; Reny and Newman Reference Reny and Newman2021; Sigelman et al. Reference Sigelman, Welch, Blesdoe and Combs1997; Tuch and Weitzer Reference Tuch and Weitzer1997). Moreover, prior research informing the FPVR model’s assumptions suggests attitudinal shifts decay with reduced salience (Birkland and Lawrence Reference Birkland and Lawrence2009; Chudy and Jefferson Reference Chudy and Jefferson2021; Nguyen et al. Reference Nguyen, Criss, Michaels, Cross, Michaels, Dwivedi and Huang2021; Tuch and Weitzer Reference Tuch and Weitzer1997). Importantly, salience is not binary. Violent Event A may be more salient than Violent Event B, but less salient than Violent Event C, such that Event A does not influence mass attitudes like Event C does. Thus, we may expect to observe an empirical pattern consistent with the dashed line in Figure 1c. H4: Initially more salient instances of civilian violence against LGBTQ+ group members will be more likely to motivate prosocial attitudes toward LGBTQ+ people than initially less salient instances of civilian violence against LGBTQ+ group members.
EVENT 1: THE PULSE MASSACRE
Studies 1 and 2 evaluate the consequences of the Pulse massacre. The massacre occurred on June 12, 2016 at the Pulse LGBTQ+ nightclub in Orlando, Florida. The massacre was perpetrated by Omar Mateen, an ISIS allegiant. Mateen killed 49 and injured 53 clubgoers with a semiautomatic rifle.Footnote 8 After taking hostages, Mateen was killed by the police. During the massacre, Pulse was hosting “Latin Night.” Eighty percent of victims were Latinx.Footnote 9
The nation reacted sympathetically post-massacre. Republican Florida Governor Rick Scott expressed support for those affected while instituting a state of emergency. The Obama administration expressed condolences and ordered federal assistance to the police investigation and the community. In a press conference, Obama described the massacre as an “act of hate.” Many on social media, including 2016 presidential election candidates, congresspeople, political figures, foreign leaders, and celebrities, expressed condolences.
The massacre was salient. 90% of adults indicated they were closely following the incident immediately post-massacre (Supplementary Figure A1). A survey that was conducted during the massacre (June 10–26) suggests that the public was aware of the shooting since it expressed more concerns about terrorism and gun violence post-massacre (Supplementary Figure A4).
Media coverage of topics related to Pulse, LGBTQ issues, and terrorism discontinuously increased post-massacre (Supplementary Figure A2). Google searches related to Pulse, LGBTQ issues, and terrorism peak when the massacre occurs (Supplementary Figure A3). Media coverage and Google searches related to these topics were either declining or limited pre-massacre, suggesting that anticipatory effects do not drive attitudinal shifts toward LGBTQ+ issues or people post-massacre. However, coverage and searches decline to their pre-incident levels by July, implying fleeting salience.
The massacre was not simply interpreted as a terror attack, but as an instance of targeted, illegitimate, anti-LGBTQ+ violence.Footnote 10 Between 70% and 85% of adults believed that the shooting was a hate crime (Supplementary Figure A5).Footnote 11 Therefore, consistent with the FPVR model, the mass public may respond prosocially to the perceptibly illegitimate Pulse massacre, given the event’s salience and concomitant sympathetic response from both the media and elites. But, given reduced media coverage and attention to the event over time, attitudinal responses may be short-lived.
Study 1: TAPS
Data and Design
Study 1 uses The American Panel Survey (TAPS, Wave 55) to assess whether exposure to violence against LGBTQ+ people motivates support for policies benefiting segments of the LGBTQ+ community. TAPS is a monthly online survey administered by the Weidenbaum Center, with national probability sampling conducted by GfK/Knowledge Networks (Weidenbaum Center 2016).
The outcome of interest is same-sex marriage support (SSM support). SSM is an important LGBTQ+ rights dimension, and it implicates multiple segments of the LGBTQ+ community. Gay, lesbian, and bisexual people who want to marry a same-sex partner benefit from legalized SSM. Transgender people who have not changed their “legal” gender but seek to marry their partner in heterosexual romantic relationships would also benefit from legalized SSM.Footnote 12 SSM approval is near-unanimous among LGBTQ+ people. Sixty percent of LGBTQ+ people say SSM should be a priority even if it takes attention from other issues.Footnote 13 TAPS asks respondents if they “generally support or oppose same-sex marriage,” with an option to indicate “no opinion.”Footnote 14 We measure SSM support as an indicator equal to 1 if the respondent indicates they support SSM, and 0 otherwise.
The independent variable is being interviewed after the Pulse massacre (post-Pulse). TAPS was fielded between June 8 and July 8, 2016. The Pulse massacre occurred on June 12, 2016, so we implement a UESD with TAPS comparing SSM support for respondents interviewed pre- and post-Pulse (Muñoz, Falcó-Gimeno, and Hernández Reference Muñoz, Falcó-Gimeno and Hernández2020). Post-Pulse is a binary indicator equal to 1 if a respondent is interviewed after June 12, 2016. Since we cannot be certain that respondents were aware of the massacre, the post-Pulse coefficient is an “intent-to-treat” (ITT) effect. However, Supplementary Figures A1–A4 suggest the public was attentive to the massacre. Moreover, TAPS respondents are more likely to believe ISIS is an important issue post-Pulse (Supplementary Figure B6), suggesting they “received the treatment” since news reports indicated that the massacre’s perpetrator had pledged fealty to ISIS. If H1 is supported, the post-Pulse coefficient would be positive.
In the absence of internal attention checks, we truncate our sample to those who completed the survey in a “reasonable duration” to account for online survey respondent inattentiveness, which may produce low-quality responses attenuating associations of interest. See DSM Section 2.4.1 for more details and evidence that this truncation does not affect our results or TAPS’ representativeness. After truncation, TAPS contains
$ N=1,142 $
respondents, 662 (58%) interviewed before Pulse and 480 after (42%).
We demonstrate that the post-Pulse coefficient is insulated from bias by validating UESD identification assumptions. The first assumption is ignorability. “Treatment” should be independent of potential outcomes conditional on random sampling. Thus, respondents interviewed pre- and post-Pulse should be compositionally similar. Figure 3a supports that assumption. Respondents interviewed post-Pulse are compositionally similar to respondents interviewed pre-Pulse across 20 baseline covariates except age (see DSM Section 2.2 for baseline covariate measurement), a finding consistent with multiple testing.
Figure 2. Fickle Prosocial Violence Response Model
Figure 3. SSM Support Increases Post-Pulse
Note: Panel (a) displays respondent covariate balance pre- and post-Pulse. Panel (b) characterizes the post-Pulse effect on SSM support with and without covariate adjustment. Panel (c) displays falsification tests characterizing the unadjusted post-Pulse effect on LGBTQ+-irrelevant outcomes. Black coefficients are statistically significant, and gray otherwise. Estimates use population weights. All covariates scaled between 0 and 1. 95% percent CIs displayed from HC2 robust SEs. See DSM Tables 91–93 for regression tables characterizing the coefficients.
Excludability is another UESD identification assumption: differences between respondents interviewed pre- and post-Pulse should be the sole consequence of the massacre. The “treatment” is not just the massacre but collateral media attention. However, other than the massacre, there are no punctuated moments of media attention to LGBTQ+ issues or violence against LGBTQ+ people during the month TAPS was fielded (June, Supplementary Figures A2 and A3), suggesting the absence of simultaneous events motivating pro-LGBTQ+ attitudes.
Additionally, it is unlikely that preexisting SSM support time trends are driving the result. To make sure, we subset TAPS to the pre-Pulse period and assess the placebo “effect” of being interviewed after the median pretreatment date and find null results (DSM Table 89).
Results
Consistent with H1, respondents interviewed post-Pulse are 13 and 10 percentage points more likely to support SSM without and with covariate adjustment (
$ p<0.05 $
; Figure 3b). These coefficients are 20%–26% of the outcome standard deviation.
We assess the robustness of our results. Our findings are likely not driven by secular dynamics outside the massacre. Falsification tests on treatment-irrelevant outcomes such as support for increasing taxes, common core, a citizenship pathway, abortion, the Keystone pipeline, ACA repeal, and emission caps are null (Figure 3c). These tests suggest chance age imbalance does not implicate balance on policy preferences.Footnote 15 Given the close association between socially conservative religious beliefs like abortion restrictionism and SSM opposition (Uecker and Froese Reference Uecker and Froese2019), the null effect of post-Pulse on abortion support in Figure 3c suggests that our results are not driven by secular shifts in social conservatism or religiosity.Footnote 16 The results are not driven by outcome item non-response since non-response is balanced pre- and post-Pulse (DSM Table 88). The results are not driven by seasonal trends; Pulse’s influence is unique to 2016. Three surveys fielded in June 2012, 2013, and 2017 show the influence of being interviewed after the massacre’s calendar day on SSM support is null (Supplementary Figure B7), suggesting no secular dynamics intrinsic to the month of June that could explain our findings (e.g., Pride Month). Our findings are robust to smaller bandwidths less susceptible to secular temporal trends (Supplementary Figure B8). Finally, given that we are deriving ITT coefficients, we test if post-Pulse is heterogeneous by political interest or news consumption. We do not find heterogeneity (Supplementary Section B.5). This is not concerning since 90% of the public was following the shooting (Supplementary Figure A1), suggesting high treatment reception regardless of dispositional political or media interest.
Temporal Persistence
Figure 4. The Influence of Pulse on SSM Support Attenuates over Time
Note: The x-axis is days cut from moment of Pulse massacre after the massacre (with days after intact). The y-axis is the post-Pulse coefficient. 95% CIs from robust SEs. See DSM Table 94 for regression table characterizing reported coefficients in this figure. See DSM Section 2.10.5 for control covariate coefficients.
We test H2 by assessing if the influence of Pulse on SSM support is temporally durable. We remove observations in the days immediately post-Pulse but not after those days, and re-analyze the influence of being surveyed post-Pulse. The logic is that respondents interviewed immediately post-Pulse may be the most susceptible to shifting attitudes toward segments of the LGBTQ+ community. Removing them may help us evaluate attitudinal decay by comparing respondents interviewed just before and some days after Pulse. After removing respondents interviewed between 1 and 10 days post-Pulse, the influence of being interviewed post-Pulse on SSM support is null (Figure 4).Footnote 17 Therefore, temporal attenuation is quick relative to prior studies demonstrating attitudinal shifts lasting several months to a year (Broockman and Kalla Reference Broockman and Kalla2016; Oskooii, Lajevardi, and Collingwood Reference Oskooii, Lajevardi and Collingwood2021). Consistent with H2, the initial SSM support increase post-Pulse was not durable.
Individual-Level Heterogeneity
We test H3a–c by assessing if the post-Pulse coefficient is larger among: (a) non-whites relative to whites and women relative to men, (b) liberals relative to moderates and conservatives, and (c) individuals living in states with a higher proportion of LGBT-identifying people and counties with a higher density of same-sex couples relative to individuals who live in areas with less LGBT-identifying people and same-sex couples.Footnote 18 Inconsistent with H3a–c, post-Pulse does not appear heterogeneous by marginalized group membership, liberalism, and LGBTQ+ geographic context (Supplementary Table B1). These findings suggest the massacre had a largely homogeneous initial influence on mass attitudes.
Study 2: PI S-IAT Data
Data and Design
Study 2 examines whether the public adopts positive attitudes toward segments of the LGBTQ+ community post-Pulse. We use Project Implicit (PI) data on U.S. respondents self-selecting into and completing an internet survey in 2016 asking questions on their explicit and implicit attitudes toward gay people via PI’s Sexuality Implicit Association Test (S-IAT,
$ N=43,950 $
) (Xu et al. Reference Xu, Nosek, Greenwald, Ratliff, Bar-Anan, Umansky and Banaji2016).Footnote 19 On average, 175 U.S. respondents completed the PI S-IAT survey daily during 2016.Footnote 20 For information on S-IAT sample composition and representativeness, see DSM Section 3.1.
The outcomes are the S-IAT D-score, straight bias, and heterocentrism. The S-IAT calculates normalized averages of how quickly respondents associate negative/positive attributes to gay/straight people relative to negative/positive attributes to straight/gay people in the form of a D-score. The D-score ranges from
$ -2 $
to
$ 2 $
. Higher values suggest implicit bias against gay people (i.e., associating negative attributes to gay people) (Greenwald and Lai Reference Greenwald and Lai2020).Footnote 21
Given indirect measurement, the D-score may be less influenced by impression management to be perceived as pro-gay post-massacre (Greenwald and Lai Reference Greenwald and Lai2020). Therefore, we can assess relatively quick, negative, emotional responses (i.e., System 2 responses) to gay people in addition to more deliberate evaluations of gay people (i.e., System 1 responses) (Greenwald and Lai Reference Greenwald and Lai2020). Although the IAT is not insulated from introspection, the modest correlation between the D-score and explicit bias suggests the IAT measures attitudes that are difficult to manipulate. Therefore, the D-score is valuable since we can demonstrate even temporary prosocial attitudinal shifts may not be impression management. The D-score is well established and associated with objective covariates characterizing subordination (Ratliff and Smith Reference Ratliff, Smith, Krosnick, Stark and ScottForthcoming).
Heterocentrism and straight bias are explicit anti-gay bias measures. Heterocentrism is the difference between respondents’ ratings on 10-point feeling thermometers for straight men and gay men. Straight bias is a 7-point measure from “I strongly prefer gay to straight people” to “I strongly prefer straight to gay people.” The D-score, straight bias, and heterocentrism are rescaled between 0 and 1.
Although heterocentrism is explicitly about gay men, and straight bias is implicitly about gay men, the D-score captures attitudes toward gay men and lesbians. In effect, the D-score measures implicit beliefs toward gay men, lesbians, and bisexuals (and transgender people in same-gender relationships). Moreover, even if our Study 2 outcomes are limited when it comes to measuring attitudes toward some members of the broader LGBTQ+ community (e.g., transgender people), attitudes toward gay people are correlated with attitudes toward transgender people (Norton and Herek Reference Norton and Herek2013), which may be pronounced given the massacre affected transgender people.Footnote 22 Therefore, our Study 2 outcomes implicate large segments of the LGBTQ+ community. Given the outcomes characterize negative attitudes, if H1 is supported, post-Pulse should be negative.
We use a UESD with the S-IAT to evaluate how anti-gay attitudes shifted post-Pulse. Given the large number of individuals taking the S-IAT daily, we estimate the influence of taking the S-IAT post-Pulse using respondents taking the S-IAT 5–50 days pre- and post-massacre in addition to the full 2016 sample between January and September. We validate the UESD ignorability identification assumption. Unlike Study 1, respondents are not sampled, but self-select, into the S-IAT. Therefore, sample composition may shift due to external events or secular trends. We expect respondents surveyed shortly pre- and post-massacre will be compositionally similar. However, respondents may be increasingly dissimilar in samples including respondents taking the survey well before or after the massacre. Supplementary Figure C10 verifies our expectation. For 5–20-day bandwidth samples (Panels A–D), there is statistical imbalance on respondent characteristics pre- and post-Pulse on 1–2/12 baseline covariates. For 25–50-day bandwidth samples, there is imbalance on 3–7 covariates (Panels E–J). Given the 15- and 20-day bandwidth samples are only imbalanced on race, we prioritize interpreting the influence of post-Pulse on anti-gay attitudes using these samples. These findings suggest our coefficient estimates, particularly for the 15- and 20-day bandwidth samples, are relatively insulated from omitted variable bias.Footnote 23
Results
Figure 5 displays post-Pulse ITT coefficients where the outcome is the D-score, straight bias, and heterocentrism. In the 15- and 20-day sample bandwidth estimates, respondents surveyed post-Pulse have a lower D-score (
$ -0.01 $
,
$ p<0.10 $
) and heterocentrism (
$ -0.01 $
,
$ p<0.01 $
), equivalent to 7% and 8% of the respective outcome standard deviations pre-Pulse. Although small, these coefficients are reasonable, likely underestimated, and substantively important vis-à-vis the target population (see DSM Section 3.4).
Figure 5. Influence of post-Pulse on Anti-Gay Attitudes
Note: The x-axis is the sample bandwidth. The y-axis is the post-Pulse coefficient. All covariates rescaled between 0 and 1. 95% CIs displayed from robust SEs. See DSM Sections 3.5.2 and 3.5.3 for corresponding regression tables.
The massacre does not appear to statistically reduce straight bias except in sample bandwidths with higher covariate imbalance (e.g., 25–50 days). Given that straight bias is highly explicit, the absence of a reliable shift in straight bias post-Pulse may be a function of impression management on part of respondents disposed against LGBTQ+ whose attitudes may otherwise shift in favor of LGBTQ+ through indirect bias measurement (Greenwald, McGhee, and Schwartz Reference Greenwald, McGhee and Schwartz1998). In sum, we find additional support for H1 in Study 2.
We conduct several robustness checks. Preexisting time trends are not driving our results (Supplementary Section C.3). We rule out if systematic temporal trends near June motivate prosocial attitudes toward gay people other than the massacre (Supplementary Section C.4). We rule out if our findings are due to a secular attitudinal trend in favor of marginalized groups (Supplementary Section C.5). We also rule out if respondent self-selection generates sorting bias (Supplementary Section C.7).
Temporal Persistence
We assess whether the D-score and heterocentrism decrease is sustainable. Consistent with H2, descriptive statistics suggest that anti-gay attitudes decreased post-Pulse, but that they rebounded to pre-Pulse levels around August (Supplementary Figure C9). We conduct a formal test of the sustainability of attitudinal shifts post-Pulse and compare S-IAT respondents surveyed 15 days pre-Pulse to those surveyed 15 days after 1–72 days post-Pulse (leaving at least 15 days up to the end of the posttreatment sample in the 2016 S-IAT data). This exercise allows us to compare individuals surveyed prior to Pulse to those surveyed some time away from Pulse at multiple time intervals. Respondents in time intervals that cut more days post-Pulse are temporally further from the massacre and potentially more subject to attitudinal decay in pro-gay beliefs. Figure 6 demonstrates the D-score and heterocentrism decrease was sustained up to 50 days post-Pulse. However, after roughly 50 days, post-Pulse attenuates toward 0.Footnote 24 Although attitudinal shifts last 50 days, these shifts are still much shorter than prior studies demonstrating long-term attitudinal shifts toward marginalized groups after external stimuli (Broockman and Kalla Reference Broockman and Kalla2016; Oskooii, Lajevardi, and Collingwood Reference Oskooii, Lajevardi and Collingwood2021). Consistent with H2, Study 2 suggests that the massacre motivated prosocial beliefs, but not durably.
Figure 6. The Influence of Post-Pulse on Reducing Anti-Gay Attitudes Attenuates over Time
Note: The x-axis is days cut from moment of Pulse massacre after the massacre (with 15 days after kept intact). The y-axis is the post-Pulse coefficient. All estimates from models adjusting for controls. 95% CIs from robust SEs. See DSM Tables 123–130 for tables characterizing the displayed coefficients. See DSM Sections 3.5.8 and 3.5.9 for control coefficients.
Individual-Level Heterogeneity
We test H3a–c by assessing if the post-Pulse coefficient is larger among non-whites, women, liberals, and individuals living in geographic contexts with more LGBTQ+ people.Footnote 25 Inconsistent with H3a–c, we find that the massacre’s influence is homogeneous. Post-pulse is not stronger for non-whites, women, liberals, or respondents in geographic contexts with more LGBTQ+ people (Supplementary Tables C4 and C5).
Mitigating Bundled Treatment Concerns
Our argument is that respondents adopted prosocial beliefs toward segments of the LGBTQ+ community after the Pulse massacre because the massacre was an instance of perceptibly anti-LGBTQ+ violence. However, the Pulse massacre was a bundled treatment in that it was also a terror attack and attack against predominantly Latinx victims. Therefore, the results in Studies 1 and 2 may be driven by the fact the Pulse massacre was a terror attack and violence against Latinxs, not just violence against LGBTQ+ people. We mitigate these concerns with several tests and evidence outlined in detail in DSM Section 1.8.
Our tests do not entirely mitigate the bundled treatment problem. Our results may be due to the combination of circumstances associated with Pulse. Therefore, we conceptually replicate Studies 1 and 2 by assessing the influence of instances of violence against LGBTQ+ group member(s) in Studies 3 and 4 that were not terror attacks nor attacks against non-whites.
EVENT 2: MATTHEW SHEPARD’S MURDER
Readers may be concerned about the external validity of Studies 1 and 2. The Pulse massacre is a unique instance of violence against LGBTQ+ people. It is the deadliest instance of violence against LGBTQ+ people, is the second deadliest mass shooting, has predominantly Latinx victims, was ISIS-inspired terrorism, and occurred after seminal gay rights victories (e.g., same-sex marriage). Therefore, it may be prudent to assess if a distinct instance of violence against LGBTQ+ group member(s) also motivates prosocial beliefs. Consequently, we examine how the murder of Matthew Shepard, a white gay Wyoming college student, by two presumptively heterosexual white men, influenced beliefs toward homosexuality during a more homophobic temporal context.
On October 6, 1998, Shepard was brutally beaten by Aaron McKinney and Russell Henderson. The incident was heavily covered by national media (Loffreda Reference Loffreda2001). Shepard died 6 days later on October 12. The murder was salient and the nation reacted sympathetically. A bipartisan group of congresspeople condemned the murder and expressed condolences. A vigil was held outside the U.S. Capitol on October 15, where thousands of people, including current and former congresspeople and celebrities, paid respects to Shepard. Advocates note Shepard’s murder engendered a “seismic shift in attitudes towards the LGBTQ community.”Footnote 26 Indeed, a decade later, Congress passed the Matthew Shepard and James Byrd Hate Crimes Prevention Act, which expanded the power to prosecute sexuality hate crimes.
On the month of Shepard’s murder, the number of gay-related news articles was 150% (NYT) and 172% (WashPo) of the January–September 1998 average (Figure 7).Footnote 27 Consistent with the FPVR model, media attention to Shepard’s murder was immediately intense but quickly declined, suggesting that attitudinal responses may be short-lived.
Figure 7. Media Coverage of Gay-Related Content in 1998
Note: Panels (a,b) display the number of NYT/Washington Post gay-related articles (y-axis) by month (x-axis). Panel (c) displays the number of gay-related articles related to Shepard or anti-gay violence by month. Dashed vertical line denotes the period Shepard is murdered.
Study 3
Data and Design
To evaluate whether Shepard’s murder decreased anti-gay attitudes, we identified surveys with similar items characterizing attitudes toward gay people shortly before and after Shepard’s murder.Footnote 28 We identify two representative CNN telephone polls asking respondents if they believe homosexuality is “morally wrong” (moral wrong) 4 months before and 2 days after Shepard’s death (CNN June 1998,
$ N=1,016 $
; CNN October 1998,
$ N=1,036 $
) (Gallup 1998; Yankelovich Partners 1998).Footnote 29 We stack these datasets and identify overlapping controls from each survey.Footnote 30 We then compare respondents interviewed after Shepard’s murder (post-Shepard) to those interviewed before the incident to assess if anti-gay violence exposure decreased the belief homosexuality is morally wrong, consistent with H1. We focus on surveys with the moral wrong outcome for three reasons. First, the question is asked on three surveys after Shepard’s murder (in 1998, 2001, and 2004), allowing an assessment of long-term attitudinal shifts. Second, there are multiple pre-Shepard surveys with the same item, allowing placebo tests to rule out if post-Shepard effects are due to secular progressive attitudinal trends concerning homosexuality’s morality. Third, moral wrong implicates large segments of the LGBTQ+ community. Lesbian, gay, bisexual, and transgender people may all partake in “homosexual” behavior. Given that the outcome characterizes a negative attitude toward segments of the LGBTQ+ community, the post-Shepard coefficient would be negative if H1 is supported.
Our approach has shortcomings we assuage. First, given the absence of auxiliary data on attention to the murder, we cannot be certain respondents “received the treatment.” Therefore, we interpret post-Shepard as an ITT effect. However, Figure 7 suggests the murder received significant media attention such that it might shift mass attitudes.
Second, given possible differences in sampling between the two surveys, our statistical conclusions may be due to sample composition. Balance tests between the two surveys demonstrate limited baseline covariate imbalance (Figure 8a), suggesting that sample composition does not drive our results.
Figure 8. Respondents Interviewed Post-Shepard Were Less Likely to Believe Homosexuality Is Morally Wrong
Note: Panel (a) characterizes balance between respondents interviewed pre- and post-Shepard’s murder. Black coefficients are significant, gray not. Panel (b) characterizes (1) the influence of being interviewed on June 1998 relative to April 1997 on the belief homosexuality is morally wrong (temporal placebo) and (2) the influence of being interviewed post-Shepard on moral wrong. Panel (c) characterizes falsification tests assessing the influence of post-Shepard on non-LGBTQ+ group attitudes. 95% CIs displayed from robust SEs. See DSM Tables 176–179 for regression tables on balance tests, the temporal placebo, the post-Shepard coefficient estimates, and falsification tests.
Third, unlike Studies 1 and 2, we cannot assess an immediate effect of anti-gay violence exposure even though the two surveys were fielded near Shepard’s murder. There are 4 months between the surveys with the moral wrong outcome (June–October 1998). Therefore, our post-Shepard estimates may be due to intervening factors or secular progressive time trends. However, there is no anti-gay violence with the level of media coverage Shepard’s murder garnered in between the field periods (Figure 7). Crowdsourced evidence suggests the last prominent instance of anti-LGBTQ+ violence prior to Shepard’s murder was not between June and October 1998, but on February 1997 (the Otherside Lounge Bombing).Footnote 31 Indeed, between June and September 1998, there were zero New York Times articles related to anti-gay hate crimes. Conversely, on the month of Shepard’s murder (October 1998), there were 17 NYT articles related to anti-gay hate crimes (Supplementary Figure D15). Two other intervening factors in 1998 may explain our results: (1) President Clinton signing an executive order against sexual orientation discrimination and (2) Tammy Baldwin’s House election (the first lesbian congressperson). We provide evidence these events are unlikely explaining our post-Shepard coefficient estimates (DSM Section 4.9 and Supplementary Section D.4).
Moreover, we rule out if our results are due to secular outcome time trends by conducting a temporal placebo test and demonstrating moral wrong levels do not change between April 1997 and June 1998 (Figure 8b).Footnote 32 These results suggest that prominent pre-study events, such as Ellen DeGeneres’s televised coming out in April 1997, are not driving our results. Despite Study 3’s shortcomings, we believe that the design provides sufficient complementary evidence to Studies 1 and 2 along with suggestive evidence our theory generalizes beyond Pulse.
Results
Consistent with H1, Figure 8b shows that respondents interviewed post-Shepard were 12 percentage points less likely to report homosexuality is morally wrong with or without covariate adjustment, 24% of the outcome standard deviation (
$ p<0.001 $
).
We conduct falsification tests on outcomes related to non-LGBTQ+ marginalized groups to rule out secular supportive trends toward marginalized groups driving our results (Figure 8c).Footnote 33 Only 4 out of 18 outcomes are significant, and the post-Shepard coefficient is not consistently in support of non-LGBTQ+ groups, suggesting no systematic secular trend driving our results (see DSM Section 4.6 for more details). Like Study 1, the null effects of post-Shepard on abortion support suggest that our results are not driven by secular shifts in social conservatism and/or religiosity.
Temporal Persistence
To assess the persistence of attitudinal shifts post-Shepard, we identify six surveys between 1978 and 2004 where the moral wrong item was asked (Harris Interactive 2004; Yankelovich Partners 1978; 1992; 2001),Footnote 34 allowing us to evaluate trends in the public’s belief that homosexuality is morally wrong pre- and post-Shepard. We do not use the CNN June 1998 poll in Figure 9 in our assessment of temporal persistence (see DSM Section 4.5 for details as to why).
Figure 9. Belief in Moral Wrong Is Stable between 1978 and 2004 with the Exception of the Moment Shepard Was Murdered
Note: Reference study is the 1994 CNN poll. Color denotes the inclusion/exclusion of controls (age, education, gender, partisanship, and race). Shaded estimate denotes Shepard’s murder (October 1998). All estimates use survey weights. All covariates scaled between 0 and 1. See DSM Table 183 for a corresponding regression table. 95% CIs displayed derived from robust SEs.
Figure 9 displays event study estimates comparing moral wrong levels in five surveys between 1978 and 2004 to a survey fielded prior to Shepard’s murder in 1994. From 1978 to 1994, belief in moral wrong is remarkably stable. Respondents surveyed in 1994 are not statistically distinct from respondents surveyed in 1992 or 1978. Consistent with our initial temporal placebo test, these findings suggest an absence of progressive attitudinal trends toward gay people prior to Shepard’s murder. However, in October 1998, immediately after Shepard’s murder, there is a statistically distinguishable decrease in moral wrong. But the decrease in the belief homosexuality is immoral is not sustainable. The mass public’s belief in the notion homosexuality is immoral returns to levels before Shepard’s murder by 2001 and 2004. Consistent with H2, our results suggest that Shepard’s murder motivated a decrease in negative beliefs concerning “homosexuals,” but this decrease was not sustainable.
Individual-Level Heterogeneity
We test H3a,b by assessing if the post-Shepard coefficient is stronger among (a) non-whites and women and (b) Democrats.Footnote 35 Given the absence of (a) county-level geographic data in the two 1998 CNN polls and (b) state-level LGBT population information in the 1990s, we cannot test H3c. We find some evidence consistent with H3a (Supplementary Table D7). Although there is no post-Shepard heterogeneity by gender, non-whites are less likely to believe homosexuality is morally wrong relative to whites post-Shepard. Whites are 7 percentage points less likely to believe homosexuality is morally wrong post-Shepard, whereas non-whites are 22 percentage points less likely, 44% of the pre-Shepard outcome standard deviation. Likewise, we find evidence supporting H3b (Supplementary Table D7). The post-Shepard effect appears driven by Democrats. Democrats are 22 percentage points less likely to believe homosexuality is morally wrong post-Shepard, whereas non-Democrats are 2 percentage points less likely.
EVENT 3: THE CLUB Q MASSACRE
Study 4 mitigates two shortcomings with Studies 1–3. First, Studies 1–3 all analyze initially highly salient events—that is, high media coverage attention. However, consistent with H4 and the FPVR model, relatively initially less salient violent events may be less likely to motivate prosocial attitudes toward segments of the LGBTQ+ community. Study 4 allows us to evaluate the consequences of indirect exposure to a putatively high-profile, but relatively initially less salient, instance of violence against LGBTQ+ group members: the 2022 Club Q massacre. Consequently, Study 4 allows us to test H4 and broader FPVR model implications related to initial event salience. Second, the outcomes in Studies 1–3 do not explicitly reference broader LGBTQ+ segments beyond gays and lesbians (e.g., transgender people). Conversely, Study 4 examines not only the same Study 2 outcomes using the 2022 PI S-IAT survey, but also additional outcomes characterizing negative attitudes toward transgender people in the 2022 PI Transgender Implicit Association Test (PI T-IAT) survey.Footnote 36 Therefore, Study 4 allows us to examine the consequences of violence against LGBTQ+ group members on mass attitudes explicitly related to transgender people, a small, politicized, population (Lewis et al. Reference Lewis, Flores, Haider-Markel, Miller and Taylor2022).
On November 19, 2022, in Colorado Springs, CO, Anderson Aldrich entered an LGBTQ+ nightclub, Club Q, and killed five clubgoers, including two trans people, while injuring 25 others with an AR-15-style rifle.Footnote 37 Aldrich was eventually incapacitated by clubgoers and apprehended by police. Evidence suggests that the violence was bias-motivated. Aldrich pleaded “no contest” in court to two hate crime charges.Footnote 38
The media and some elites reacted sympathetically to the violence. President Biden and Transportation Secretary Buttigieg immediately expressed condolences.Footnote 39 However, unlike the Pulse massacre and Shepard’s murder, the elite response was relatively polarized. Buttigieg blamed the shooting on growing Republican anti-LGBTQ+ rhetoric.Footnote 40 Tucker Carlson and several right-wing commentators blamed the violence on purported “grooming” activity from LGBTQ+ people.Footnote 41 Republican politicians who expressed condolences were criticized for simultaneously engaging in anti-LGBTQ+ rhetoric.Footnote 42 LGBTQ+ advocates noted a rise in queerphobic posts across social media platforms post-shooting.Footnote 43
Moreover, relative to Shepard’s murder and the Pulse massacre, the Club Q massacre was less salient. First, there were less NYT articles related to the Club Q massacre 2 months after the event relative to Shepard’s murder and the Pulse massacre (Supplementary Figure E18). Second, regression discontinuity-in-time estimates suggest that although online articles on topics related to mass shootings, the LGBT community, and hate crimes discontinuously increased after Club Q, there were more online articles on topics related to mass shootings and the LGBT community after Pulse (Supplementary Figures E19 and E20 and Supplementary Table E8). Third, Google search data demonstrate that there was more attention to mass shootings, LGBT people, and LGBT hate crimes immediately during Pulse relative to immediately during the Club Q massacre (Supplementary Figure E21). Therefore, consistent with the FPVR model and H4, although Club Q was relatively high-profile, its lower-profile status vis-à-vis Pulse and Shepard’s murder suggests that it may be less likely to initially shift mass attitudes.
Study 4
Data and Design
We use data on U.S. respondents self-selecting into the 2022 PI S-IAT (
$ N=184,824 $
, 506 daily average respondents) and T-IAT (
$ N=85,303 $
, 233 daily average respondents) surveys. See DSM Section 5.1 for information on S-IAT and T-IAT sample composition and representativeness (Xu et al. Reference Xu, Nosek, Greenwald, Ratliff, Bar-Anan, Umansky and Banaji2022a; Reference Xu, Nosek, Greenwald, Ratliff, Bar-Anan, Umansky and Banaji2022b).
The S-IAT outcomes are the same as Study 2’s (anti-gay D-score, heterocentrism, and straight bias). The three T-IAT outcomes are similar but slightly different. The anti-trans D-score is measured by assessing the speed by which respondents associate negative/positive attributes (words) to images of trans/cis celebrities. Higher values suggest respondents associated negative/positive attributes to trans people faster/slower than cis people. Ciscentrism measures relative warmth toward cisgender people vis-à-vis trans people. Cis bias is a 7-point scale measuring preferences for cisgender relative to trans people. See DSM Section 5.2 for more T-IAT outcome measurement details. Prior research finds that the T-IAT outcomes are correlated with anti-trans policy preferences (Axt et al. Reference Axt, Conway, Westgate and Buttrick2021). All outcomes are rescaled between 0 and 1.
The main independent variable is post-Club Q, an indicator equal to 1 if a respondent self-selects into the S-IAT or T-IAT after November 19, 2022. The post-Club Q coefficients will be negative if prosocial attitudes increase post-Club Q.
We implement another UESD, estimating the influence of post-Club Q 5–40 days in 5-day intervals post-massacre.Footnote 44 We assess covariate balance for these bandwidth samples between respondents taking the S-IAT/T-IAT pre- and post-Club Q (Supplementary Figures E22 and E23).Footnote 45 Covariate imbalance increases as sample bandwidth increases, likely due to unobservable secular trends. Therefore, we primarily interpret the 20- and 15-day bandwidth samples in the S-IAT and T-IAT, respectively, where there is the least imbalance (4/12 and 1/12 covariates imbalanced, respectively).
Results
The post-Club Q coefficient is null across all outcomes in the 20-day bandwidth sample for the S-IAT and in the 15-day bandwidth sample for the T-IAT (Figure 10). Although post-Club Q coefficients in larger bandwidth samples suggest a decrease in the anti-trans and anti-gay D-score (e.g., the 40-day bandwidth samples), these estimates should be viewed skeptically, given that they possess high covariate imbalance and are more likely to be perturbed by unobservable secular trends (Supplementary Figures E22 and E23). Consistent with the FPVR model and H4, less salient violent events like Club Q do not motivate attitudinal shifts like more salient events (e.g., Pulse or Shepard’s murder).
Figure 10. Influence of Post-Club Q on Anti-Gay, Anti-Trans Attitudes
Note: The x-axis is the bandwidth sample (1–40 days). The y-axis is the post-Club Q coefficient. Panels characterize different outcomes. The top/bottom three panels characterize estimates from the 2022 PI S-IAT/T-IAT data. Black coefficients are from models adjusting for controls, gray otherwise. 95% CIs displayed from HC2 robust SEs. See DSM Tables 185 and 186 for regression tables characterizing these estimates.
Individual-Level Heterogeneity
We test H3a–c and assess if the post-Club Q coefficient is larger among (a) non-whites and women, (b) liberals, and (c) individuals living in geographic contexts with more LGBTQ+ group members using the 20- and 15-day bandwidth samples for the S-IAT and T-IAT. We find limited heterogeneity across these characteristics (see Supplementary Tables E9 and E10). The only statistically significant heterogeneity we identify is that the post-Club Q coefficient is negative and stronger among women for the Cis Bias outcome (Supplementary Table E10). However, we do not identify heterogeneity by gender in the S-IAT data or the other two T-IAT outcomes. Therefore, we interpret the influence of post-Club Q as largely homogeneous.
Evidence from Less Salient Violent Events
A limitation with Study 4 is that although the Club Q massacre was less salient than Pulse and Shepard’s murder, the null results may be due to the arguably more polarized temporal context given the recent rise of Republican anti-LGBTQ+ rhetoric and policies. Anti-LGBTQ+ laws implemented in Republican states (Supplementary Figure E24) and right-wing anti-LGBTQ+ protests have increased in the past few years (Supplementary Figure E25). Indeed, prior research shows LGBTQ+ mass attitudes may entrench in polarized contexts (Lewis et al. Reference Lewis, Flores, Haider-Markel, Miller and Taylor2022). The FPVR model also corroborates this limitation, since sympathetic responses by bipartisan elites may be necessary to motivate prosocial mass attitudes (Figure 2).
To circumvent this limitation, we use crowdsourced data on less salient violent incidents against LGBTQ+ people between 2010 and 2022 and evaluate the influence of these events on prosocial attitudes.Footnote 46 We demonstrate the incidents outside of those in Studies 1–4 are significantly less salient (Supplementary Figure F26). We identify 3,570, 442, and 358 NYT article hits related to the Pulse massacre, Shepard’s murder, and the Club Q massacre, respectively (Supplementary Figure F26c). Conversely, the next most salient violent incident against LGBTQ+ group members between 2010 and 2022 was Mark Carson’s May 2013 murder with 30 hits (Figure F26b). Consistent with H4, other less salient violent incidents against LGBTQ+ group members outside those in Studies 1–4 have largely null effects on mass attitudes toward gay people (Supplementary Figure F27). The few significant effects are not consistently in the same substantive direction, implying a random, unsystematic, causal process.
LIMITATIONS AND ADDITIONAL ROBUSTNESS CHECKS
Our analyses have limitations. First, one issue with our analytic approach is that we use several distinct outcomes across different time periods while assuming they measure the same concept (i.e., prosocial LGBTQ+ attitudes). We show this is an advantage, rather than a shortcoming, in DSM Section 7.
Second, although we provide evidence respondents likely perceived and responded to violence against LGBTQ+ people in a manner consistent with the FPVR model, we cannot be certain respondents “received the treatment.” Future research should use designs encouraging stronger treatment reception (e.g., survey experiments) to assess if our analyses underestimate effects and/or temporal persistence. However, unlike designs offering stronger treatment reception, a (tragic) advantage of our design(s) is that they derive effects based on “real-world,” externally valid events.
Third, our evidence has not tested all FPVR model mechanisms. Our design is advantageous in that we can assess the effects of violence on prosocial attitudes in an uncontrolled environment with plausible identification assumptions, undercutting demand effects or external invalidity. But, our data were not directly collected to test our hypotheses, making mechanism tests difficult. To the extent we can provide evidence for FPVR model mechanisms (Figure 2), we show that initial salience is necessary to motivate prosocial attitudes at the outset, that declines in salience over time are concomitant with decay in prosocial attitudinal shifts, and that there is limited support shared marginalization, ideology, and LGBTQ+ geographic context consistently moderate the initial adoption of prosocial attitudes.
Future research should test other FPVR model mechanisms (Figure 2). Psychological insights are promising. Violence exposure’s influence on prosocial beliefs and their sustainability may be mediated through positive emotional responses toward marginalized groups (e.g., empathy, sympathy, anger, and guilt) (Branscombe and Miron Reference Branscombe, Miron, Tiedens and Leach2004; Harth, Kessler, and Leach Reference Harth, Kessler and Leach2008). Additionally, future research should assess how media frames condition the public’s attitudinal responses. During Shepard’s murder and Pulse, the media and elites framed the victims sympathetically (instead of unsympathetically). Concomitantly, prior research suggests the media used episodic frames focusing on perpetrator motivations instead of thematic frames emphasizing societal queerphobia (Ott and Aoki Reference Ott and Aoki2002; Zahzah Reference Zahzah2019). Framing differences may condition prosocial responses and their temporal durability.
Fourth, another limitation is that we only focus on indirect exposure to high-profile violence. Direct observation of smaller-scale quotidian violence against LGBTQ+ group members (e.g., observing hate crimes, assault, and verbal abuse) may have a stronger, durable influence on prosocial beliefs. Future research should explore how different violence exposure types motivate prosocial beliefs.
Fifth, another limitation is that we only explore attitudinal shifts, not behavior. See DSM Section 1.7 for reasoning and evidence the lack of behavioral emphasis may not be a shortcoming.
CONCLUSION
We present an FPVR model to explain how indirect exposure to civilian violence against marginalized groups may influence prosocial attitudes toward targeted groups. Across four studies and three events, we provide evidence supporting the model and show indirect civilian violence against LGBTQ+ group members increases prosocial attitudes toward segments of the LGBTQ+ community. However, these prosocial responses are not temporally sustainable and less salient events do not motivate prosociality at the outset. Our core contribution is that we repeatedly demonstrate indirect exposure to salient civilian violence against marginalized groups may not sustainably undercut negative attitudes toward these groups. The FPVR model provides a general framework that can be tested and theoretically built upon in domains outside anti-LGBTQ+ violence, such as violence against other marginalized groups (e.g., non-whites, immigrants, and women).
Interestingly, we find limited individual-level heterogeneity in Studies 1, 2, and 4,Footnote 47 and some evidence non-whites and Democrats are more likely to adopt prosocial attitudes after Shepard’s murder in Study 3. The absence of heterogeneous effects in Studies 1 and 2 are not necessarily surprising. The Parallel Publics thesis posits salient events can generate common information exposure and therefore homogeneous attitudinal responses across population subgroups (Page and Shapiro Reference Page and Shapiro2010). Relatedly, there was mainstream agreement among media and elites the Pulse massacre was tragic and reflected illegitimate behavior. Thus, messaging associated with the massacre was not a “group cue” that could motivate prosocial responses among some subgroups but not others (Zaller Reference Zaller1992). Indeed, the effect homogeneity we identify is consistent with prior evidence showing SSM support moves in parallel over time across partisan and social subgroups (Coppock Reference Coppock2023). Study 3’s individual-level heterogeneity may be a function of temporal context. Relative to 2016, racial violence was salient in 1998. James Byrd was murdered 4 months before Shepard’s murder. The Rampart LAPD scandal was also underway (involving the police beating of Ishmael Jimenez). Therefore, non-whites may have been primed to adopt prosocial attitudes toward groups facing conceivably analogous violence. Likewise, the mass public was less acceptant toward LGBTQ+ people in the 1990s. Therefore, socially conservative Republicans and independents may have been resistant to sympathetic messaging after Shepard’s murder relative to liberal Democrats. Finally, Study 4’s limited heterogeneity may be due to Club Q’s limited salience vis-à-vis Pulse and Shepard’s murder.
What would generate durable effects? The FPVR model suggests sustained media attention may motivate sustained attitudinal shifts (Figure 2). Disturbingly, salient violent event recurrence may facilitate sustainable prosocial shifts. Additionally, the FPVR model posits elites play a role in making violent incidents salient. Therefore, elites who continue to strategically amplify issues related to a specific event long after occurrence may sustain attitudinal shifts (Birkland Reference Birkland1998; Zaller Reference Zaller1992). The masses may also play a role in facilitating continued event salience. Reny and Newman (Reference Reny and Newman2021) show prosocial attitudinal responses to anti-Black violence are relatively durable if the violence is concomitant with (a very large and sustained) social protest. Moreover, perhaps direct or proximal, as opposed to indirect, violence exposure is necessary to durably shift mass attitudes, consistent with prior work (Hadzic, Carlson, and Tavits Reference Hadzic, Carlson and Tavits2020; Lupu and Peisakhin Reference Lupu and Peisakhin2017; Mironova and Whitt Reference Mironova and Whitt2018). The FPVR model could also be extended by evaluating effect sustainability conditional on victim or perpetrator characteristics (e.g., state- vs. civilian-perpetrated), and the scale of violence. We leave it to future research to continue to develop new theoretical insights, extend the FPVR model, and assess possibilities for durable effects.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit https://doi.org/10.1017/S0003055424000947.
DATA AVAILABILITY STATEMENT
Research documentation and data that support the findings of this study are openly available at the American Political Science Review Dataverse: https://doi.org/10.7910/DVN/PA2XXO.
ACKNOWLEDGEMENTS
We thank Roberto Carlos, Kiela Crabtree, Donald Grasse, Wayde Marsh, Efrén Pérez, Michael Shepherd, participants at the University of Texas–Austin Politics of Race and Ethnicity (PRE) lab, the four anonymous reviewers, and the APSR editors for their insightful and helpful feedback.
CONFLICT OF INTEREST
The authors declare no ethical issues or conflicts of interest in this research.
ETHICAL STANDARDS
The authors affirm this research did not involve human participants.
Comments
No Comments have been published for this article.