BACKGROUND:

Bias-based bullying is associated with negative outcomes for youth, but its contextual predictors are largely unknown. Voter referenda that target lesbian, gay, bisexual, and transgender groups may be 1 contextual factor contributing to homophobic bullying.

METHODS:

Data come from 14 consecutive waves (2001–2014) of cross-sectional surveys of students participating in the California Healthy Kids Survey (N = 4 977 557). Student responses were aggregated to the school level (n = 5121). Using a quasi-experimental design, we compared rates of homophobic bullying before and after Proposition 8, a voter referendum that restricted marriage to heterosexuals in November 2008.

RESULTS:

Interrupted time series analyses confirmed that the academic year 2008–2009, during which Proposition 8 was passed, served as a turning point in homophobic bullying. The rate of homophobic bullying increased (blinear = 1.15; P < .001) and accelerated (bquadratic = 0.08; P < .001) in the period before Proposition 8. After Proposition 8, homophobic bullying gradually decreased (blinear = −0.28; P < .05). Specificity analyses showed that these trends were not observed among students who reported that they were bullied because of their race and/or ethnicity, religion, or gender but not because of their sexual orientation. Furthermore, the presence of a protective factor specific to school contexts among lesbian, gay, bisexual, and transgender youth (gay-straight alliances) was associated with a smaller increase in homophobic bullying pre–Proposition 8.

CONCLUSIONS:

This research provides some of the first empirical evidence that public campaigns that promote stigma may confer risk for bias-based bullying among youth.

What’s Known on This Subject:

Bias-based bullying is common among youth and is associated with adverse outcomes, but the predictors of bias-based bullying, especially at the social-ecological level, are largely unknown.

What This Study Adds:

We documented accelerated rates of homophobic bullying in California during a voter referendum that restricted marriage to heterosexuals. This research provides some of the first empirical evidence that voter referenda that promote stigma confer risk for bias-based bullying among youth.

Bullying is a widespread form of peer aggression1,2 associated with numerous deleterious psychosocial and health outcomes among youth that can persist into adulthood.3,4 Research has recently focused on a specific form of bullying known as bias-based bullying, which is motivated by a victim’s actual or perceived membership in a minority, often stigmatized, group. Bias-based bullying is common among youth (with some estimates as high as 40%)5 and is more strongly associated with adverse outcomes than bullying unrelated to bias.5,7 However, few studies have identified predictors of bias-based bullying, particularly at the contextual and/or ecological levels.8 

Our study addresses this important gap by examining whether 1 contextual and/or ecological factor, a voter referendum that promulgated stigma against a particular group, contributes to bias-based bullying. We focused in particular on whether Proposition 8, a California voter referendum in November 2008 that restricted marriage to heterosexuals,9 was associated with trends in homophobic bullying (bullying related to actual or perceived sexual orientation) among California youth. Although Proposition 8 was passed in November 2008, it was preceded and immediately followed by numerous legal activities as well as associated media and public campaigns (eg, responses to the proposition by the California State Senate and Supreme Court of California in the spring of 2009).10 Thus, we hypothesized a quadratic functional form to trends in homophobic bullying, with the 2008–2009 academic school year serving as an inflection point; specifically, we expected that the rate of homophobic bullying would (1) increase and accelerate in the period leading up to the vote on Proposition 8 (ie, 2008–2009 school year) and (2) gradually decline in subsequent years after a decrease in Proposition 8–related activities. This hypothesis is consistent with previous work demonstrating that voter campaigns that target lesbian, gay, bisexual, and transgender (LGBT) groups create a social environment that is conducive to bias against sexual and gender minorities,11,14 which may manifest in heightened levels of victimization against LGBT youth.

To strengthen the inferences that we were able to draw about the association between Proposition 8 and homophobic bullying, we test 2 possible alternative explanations. The first examined whether Proposition 8 was associated with trends in bias-based bullying among students who reported that they were bullied because of their race and/or ethnicity, religion, or gender but not their sexual orientation. Such an analysis is considered a negative control approach15 in that we test whether there is an association among groups in which we would not theoretically expect it. Our confidence that the association is specific to Proposition 8 would be bolstered if we found that it was associated with the trajectory of homophobic bullying but not of the other 3 types. Second, we determined if the presence of gay-straight alliances (GSAs) moderated the relationship between Proposition 8 and homophobic bullying. GSAs are school-based clubs focused on improving the school climate for LGBT youth, and a large literature documents that GSAs protect LGBT youth against mental health problems and victimization.16,17 We examined whether the presence of GSAs was associated with smaller increases in homophobic bullying before the Proposition 8 vote. If the trends in homophobic bullying revealed by our analysis were due to some unmeasured historical event co-occurring with Proposition 8, it seems unlikely that we would find evidence that GSAs, a factor specific to school contexts among LGBT youth, serve a protective role against changes in rates of homophobic bullying.

To evaluate these hypotheses, we strategically combine data from nearly 5 million youth from >5000 schools in California across 14 school years linked to statewide data on GSAs in California. In 2000, California passed a law, the California Student Safety and Violence Prevention Act (AB 537), that changed the state’s Education Code by adding actual or perceived sexual orientation and gender identity to the existing nondiscrimination policy. Thus, we are able to take advantage of a rare opportunity to evaluate our research questions in a state that had already passed an inclusive antidiscrimination policy before the study period began.

We obtained data from the California Healthy Kids Survey (CHKS), the largest statewide survey in the United States of youth risk behaviors and protective factors. Most secondary schools in California participated in the study once every 2 academic years, whereas a small percentage of schools participated annually. Schools participating biennially versus annually did not differ in rates of homophobic bullying at any wave (Supplemental Information); consequently, all schools were included in the analyses. Average student response rates were typically >70% (eg, 72% and 71% in 2011–2013 and 2013–2015, respectively).18 We used 14 consecutive waves of data collected between the 2001–2002 and the 2014–2015 academic years, which include the months of September to June in each year (N = 4 977 557 students). The study was reviewed by the Institutional Review Board at The University of Texas at Austin and exempted because deidentified data were obtained from secondary sources.

Homophobic Bullying

Homophobic bullying was measured via the question, “During the past 12 months, how many times on school property were you harassed or bullied because you are gay or lesbian (or someone thought you were)?” Response options included the following: never, once, 2 to 3 times, or 4 or more times. Responses were highly left skewed, with most students indicating no homophobic bullying. Thus, a dichotomous measure was created (0 = “not bullied”; 1 = “bullied”) for each wave. Rates of homophobic bullying were aggregated at the school level (n = 5121) for each academic year and calculated by using the number of students reporting homophobic bullying divided by the number of total students in each school for each academic year. All schools that participated within a particular survey year were included in this calculation irrespective of their previous or future participation in the CHKS. Sensitivity analyses revealed that spring reports of homophobic bullying were higher than fall reports for each available year of data on the timing of survey assessment (fall 2001–spring 2009). However, the overall trend for fall and spring semesters is nearly identical to the trends observed when using average rates across the academic year (Supplemental Information); consequently, data were averaged across both semesters.

Other Types of Bias-Based Bullying

Students were asked how often they had been bullied or harassed on school property during the past 12 months because of their “race, ethnicity, or national origin”; religion; or gender. These variables were recoded into dichotomous indicators of bias-based bullying (0 = “not bullied”; 1 = “bullied”) for each of these 3 types.

GSAs in Schools

Dichotomous variables indicating whether a school had a GSA for each year were created on the basis of annual data of GSA members’ registration of their GSAs in California schools with the GSA Network (a statewide education and advocacy organization supporting school GSAs). GSAs are modeled as a time-varying predictor because we had information on the school year during the study period in which the GSA was registered.

Proposition 8 Referendum

Proposition 8 was passed in November 2008. Thus, we examine homophobic bullying trends before and after the 2008–2009 academic year, which includes the month in which Proposition 8 voting occurred.

To test our study hypotheses, we examined school-level trends in rates of homophobic bullying in California schools before and after Proposition 8 voting via interrupted time series analysis (ITS),19,20 a statistical tool for assessing associations between policy or legislation and outcomes of interest in nonexperimental data. Within a time series of repeated observations (in this case, 14 academic years from 2001–2002 to 2014–2015), ITS compares the rates of a phenomenon (in this case, homophobic bullying) before and after a policy or legislative change (in this case, the vote on Proposition 8).20 Segmented regression, a model with different intercept and slope coefficients for the pre- and postlegislation period, was used to estimate both the level (mean) and trend (slope) of homophobic bullying. A notable strength of ITS for our purposes is that secular trends in homophobic bullying are controlled for. ITS thus allows for examination of level and slope changes associated with Proposition 8 voting while controlling for the overall trend in homophobic bullying over time.20 

Although there is little agreement with respect to the exact number of time points needed before and after the event of interest, the analytical power of ITS is typically stronger with larger numbers of time points. In situations in which the data series are balanced before and after the change of interest, adequate power can be achieved with as few as 12 time points.20,21 We have 14 relatively balanced time points (7 before Proposition 8 voting and 6 after the vote), indicating adequate power for ITS.

We began by analyzing the simplest model, testing a baseline mean model (ie, no systematic increase or decrease over time) followed by increasingly more complex linear and quadratic models. These analyses revealed a significant quadratic (curvilinear) trend in homophobic bullying before the vote. Thus, we conducted a mixed-effect quadratic ITS, estimating variation in rates of homophobic bullying over time between schools (the school random effect) as well as linear and quadratic trends in homophobic bullying over time (fixed effects). We included a prevote-versus-postvote binary indicator to compare the prevote (2001–2002 through 2007–2008; coded 0) versus postvote (2008–2009 through 2014–2015; coded 1) periods (fixed effect). Finally, interactions between the prevote-versus-postvote indicator and linear and quadratic trend terms, respectively, were included to examine whether postvote linear and quadratic trends in homophobic bullying differed significantly relative to their respective predicted prevote trends (fixed effects). Because time (academic year) is nested within schools, we estimated robust SE clustering by school. By accounting for differing baseline school rates of homophobic bullying (ie, random intercepts), the mixed-effect model captures relative changes within schools while controlling for time-invariant school characteristics.

To examine GSAs as a moderator of the relationship between Proposition 8 and homophobic bullying, we added the school-level GSA indicator (0 = no GSA; 1 = GSA) to the basic model described above as well as 2-way interaction terms between GSA and the linear trend in homophobic bullying, GSA and the quadratic trend, and GSA and the prevote-versus-postvote indicator. Finally, we included 3-way interactions between GSA, the prevote-versus-postvote indicator, and the linear and quadratic trends to assess whether schools with versus without GSAs differed regarding changes in linear and quadratic trends in homophobic bullying, respectively, before versus after Proposition 8 voting. Significant interactions were probed to obtain linear and quadratic trends (slopes) after the vote. Additional details on the ITS, including the equations used in the models, are provided in the Supplemental Information.

Preliminary analyses (Supplemental Information) indicated that the inclusion of several school-level factors (eg, racial and/or ethnic diversity, socioeconomic status, and school size) did not change the direction or magnitude of the results and thus were not included as controls. Missing data were minimal for the homophobic bullying item (7.37%); thus, missing data were handled by using listwise deletion. All analyses were conducted by using Stata 14 (Stata Corp, College Station, TX).22 

Descriptive statistics regarding the rates of homophobic bullying and the presence of school GSAs for each of the 14 academic years are presented in Table 1.

TABLE 1

Rates of Homophobic Bullying and GSAs in California Schools From 2001 to 2014

Academic YearHomophobic Bullying, %GSA, %
2001–2002 7.6 2.7 
2002–2003 8.1 4.7 
2003–2004 8.3 6.3 
2004–2005 8.9 7.1 
2005–2006 9.1 8.2 
2006–2007 9.6 9.2 
2007–2008 10.6 10.5 
2008–2009a 10.8 11.4 
2009–2010 10.4 12.1 
2010–2011 10.3 13.2 
2011–2012 10.1 14.7 
2012–2013 9.9 15.1 
2013–2014 9.8 16.1 
2014–2015 9.2 16.2 
Academic YearHomophobic Bullying, %GSA, %
2001–2002 7.6 2.7 
2002–2003 8.1 4.7 
2003–2004 8.3 6.3 
2004–2005 8.9 7.1 
2005–2006 9.1 8.2 
2006–2007 9.6 9.2 
2007–2008 10.6 10.5 
2008–2009a 10.8 11.4 
2009–2010 10.4 12.1 
2010–2011 10.3 13.2 
2011–2012 10.1 14.7 
2012–2013 9.9 15.1 
2013–2014 9.8 16.1 
2014–2015 9.2 16.2 
a

Proposition 8 voting took place in November of the 2008–2009 academic year. Rates of homophobic bullying were calculated on the basis of the student-level data (N = 4 977 557).

The rate of homophobic bullying during the prevote academic years from 2001–2002 to 2007–2008 increased (blinear = 1.15; P < .001) and accelerated (bquadratic = 0.08; P < .001) in the years before the Proposition 8 vote (Table 2). Additionally, the interaction terms for the binary prevote-versus-postvote indicator and quadratic and linear trends, respectively, were both significant, indicating that the 2008–2009 academic year (when Proposition 8 was passed) was an inflection point in rates of homophobic bullying. After the vote, homophobic bullying gradually decreased (blinear = −0.28; P < .05).

TABLE 2

Trends in Homophobic Bullying in California Schools Before and After the Vote on Proposition 8

Fixed EffectsbSE95% CI
Prevote trends    
 Linear  1.15*** 0.13 0.90 to 1.41 
 Quadratic  0.08*** 0.02 0.05 to 0.11 
Pre- versus postvote change    
 Postvote −0.71* 0.28 1.27 to 0.15 
 Linear trend*postvote −1.44*** 0.18 1.80 to 1.08 
 Quadratic trend*postvote −0.09*** 0.03 0.14 to 0.04 
Probed postvote trends    
 Linear  −0.28* 0.13 0.54 to 0.03 
 Quadratic  −0.01 0.02 0.05 to 0.03 
Random effect    
 School mean 13.09a 2.08 9.58 to 17.88 
 Residual 38.70a 1.99 35.00 to 42.80 
Fixed EffectsbSE95% CI
Prevote trends    
 Linear  1.15*** 0.13 0.90 to 1.41 
 Quadratic  0.08*** 0.02 0.05 to 0.11 
Pre- versus postvote change    
 Postvote −0.71* 0.28 1.27 to 0.15 
 Linear trend*postvote −1.44*** 0.18 1.80 to 1.08 
 Quadratic trend*postvote −0.09*** 0.03 0.14 to 0.04 
Probed postvote trends    
 Linear  −0.28* 0.13 0.54 to 0.03 
 Quadratic  −0.01 0.02 0.05 to 0.03 
Random effect    
 School mean 13.09a 2.08 9.58 to 17.88 
 Residual 38.70a 1.99 35.00 to 42.80 

Variance explained for the model is 0.036. Rates of homophobic bullying were estimated on the basis of the aggregated school-level data (n = 5121). Time was centered on the 2008–2009 academic year (postvote = 0 [2001–2002 through 2007–2008] versus 1 [2008–2009 through 2014–2015]). The coefficient for postvote indicates the prevote-versus-postvote intercept change. The coefficient for linear trend*postvote indicates the prevote-versus-postvote change in the linear trend; the coefficient for quadratic trend*postvote indicates the prevote-versus-postvote change in the quadratic trend. Postvote linear and quadratic trends were probed via post hoc analyses of significant interactions. CI, confidence interval.

a

Value refers to variance.

*

P < .05.

***

P < .001.

As shown in Table 3 and Fig 1, schools with a GSA had a smaller increase in rates of homophobic bullying in the pre–Proposition 8 period (bintercept = −2.50 [P < .001]; blinear = −0.58 [P < .01]; bquadratic = −0.06 [P < .05]) compared with schools without a GSA. The 3-way interaction term comparing differences in the change of linear trends in schools with versus without a GSA was also significant (b = 1.00; P < .001), indicating that the trend change in homophobic bullying pre–Proposition 8 to post–Proposition 8 was less dramatic for schools with a GSA than those without.

TABLE 3

The Moderating Role of School GSAs on Rates of Homophobic Bullying in California Schools Before and After the Vote on Proposition 8

Fixed EffectsbSE95% CI
Prevote    
 Reference: schools without GSAs    
  Linear trend 1.26*** 0.15 0.97 to 1.54 
  Quadratic trend 0.09*** 0.02 0.05 to 0.12 
 GSA effect    
  GSA −2.50*** 0.38 −3.24 to −1.76 
  Linear trend*GSA −0.58** 0.21 −1.00 to −0.16 
  Quadratic trend*GSA −0.06* 0.03 −0.11 to −0.004 
Prevote-versus-postvote change    
 Reference: schools without GSAs    
  Postvote −0.71* 0.33 −1.35 to −0.07 
  Linear trend*postvote −1.59*** 0.21 −2.01 to −1.18 
  Quadratic trend*postvote −0.10** 0.03 −0.15 to −0.04 
 GSA effect    
  Postvote*GSA 0.10 0.45 −0.79 to 0.98 
  Linear trend*postvote*GSA 1.00*** 0.31 0.38 to 1.62 
  Quadratic trend*postvote*GSA 0.03 0.05 −0.07 to 0.13 
Probed postvote trends    
 Schools without GSAs    
  Linear trend −0.33* 0.15 −0.63 to −0.03 
  Quadratic trend −0.01 0.02 −0.06 to 0.04 
 Schools with GSAs    
  Linear trend 0.09 0.17 −0.26 to 0.43 
  Quadratic trend −0.04 0.04 −0.11 to 0.03 
Random effect    
 School mean 10.89a 1.70 8.01 to 14.79 
 Residual 38.57a 1.95 34.94 to 42.58 
Fixed EffectsbSE95% CI
Prevote    
 Reference: schools without GSAs    
  Linear trend 1.26*** 0.15 0.97 to 1.54 
  Quadratic trend 0.09*** 0.02 0.05 to 0.12 
 GSA effect    
  GSA −2.50*** 0.38 −3.24 to −1.76 
  Linear trend*GSA −0.58** 0.21 −1.00 to −0.16 
  Quadratic trend*GSA −0.06* 0.03 −0.11 to −0.004 
Prevote-versus-postvote change    
 Reference: schools without GSAs    
  Postvote −0.71* 0.33 −1.35 to −0.07 
  Linear trend*postvote −1.59*** 0.21 −2.01 to −1.18 
  Quadratic trend*postvote −0.10** 0.03 −0.15 to −0.04 
 GSA effect    
  Postvote*GSA 0.10 0.45 −0.79 to 0.98 
  Linear trend*postvote*GSA 1.00*** 0.31 0.38 to 1.62 
  Quadratic trend*postvote*GSA 0.03 0.05 −0.07 to 0.13 
Probed postvote trends    
 Schools without GSAs    
  Linear trend −0.33* 0.15 −0.63 to −0.03 
  Quadratic trend −0.01 0.02 −0.06 to 0.04 
 Schools with GSAs    
  Linear trend 0.09 0.17 −0.26 to 0.43 
  Quadratic trend −0.04 0.04 −0.11 to 0.03 
Random effect    
 School mean 10.89a 1.70 8.01 to 14.79 
 Residual 38.57a 1.95 34.94 to 42.58 

Variance explained for the model is 0.058. Rates of homophobic bullying were estimated on the basis of the aggregated school-level data (n = 5121). Time was centered on the 2008–2009 academic year (postvote = 0 [2001–2002 through 2007–2008] versus 1 [2008–2009 through 2014–2015]). GSA = 0 [for schools with no GSAs] versus 1 [for schools with GSAs]. Coefficients that involve GSAs indicate the differences in schools with GSAs versus schools without GSAs in intercept, linear slope, and quadratic slope. Coefficients that involve the postvote indicate the prevote-versus-postvote changes in intercept, linear slope, and quadratic slope. Postvote linear and quadratic trends were probed via post hoc analyses of significant interactions. CI, confidence interval.

a

Value refers to variance.

*

P < .05.

**

P < .01.

***

P < .001.

FIGURE 1

The moderating role of school GSAs on rates of homophobic bullying in California schools before and after the vote on Proposition 8. Rates of homophobic bullying were estimated on the basis of the aggregated school-level data (n = 5121). The vertical line indicates when Proposition 8 voting took place (November of the 2008–2009 academic year).

FIGURE 1

The moderating role of school GSAs on rates of homophobic bullying in California schools before and after the vote on Proposition 8. Rates of homophobic bullying were estimated on the basis of the aggregated school-level data (n = 5121). The vertical line indicates when Proposition 8 voting took place (November of the 2008–2009 academic year).

Close modal

We next conducted analyses among students who reported that they were bullied because of their race and/or ethnicity, religion, or gender but not because of their sexual orientation. Rates of bias-based bullying due to race and/or ethnicity (blinear = −1.04 [P < .001]; bquadratic = −0.18 [P < .001]), religion (bquadratic = −0.03; P < .05), and gender (blinear = −0.11; P < .001) all decreased before Proposition 8 (Table 4); the trends were in the opposite direction for those reporting homophobic bullying (Table 2). Additional specificity analyses (Supplemental Information) provide additional evidence that the results were concentrated among those who only reported homophobic bullying.

TABLE 4

ITS Analysis for Students Reporting Any Type of Bias-Based Bullying Other Than Homophobic Bullying: California Students (2001–2014)

Fixed EffectsRace and/or EthnicityReligionGender
bSEbSEbSE
Prevote       
 Linear trend −1.04*** 0.13 −0.16 0.10 −0.11*** 0.02 
 Quadratic trend −0.18*** 0.02 −0.03* 0.01 — — 
Prevote-versus-postvote change       
 Postvote 0.54* 0.26 −0.13 0.21 −0.48*** 0.12 
 Linear trend*postvote 0.70*** 0.17 −0.08 0.13 −0.10** 0.03 
 Quadratic trend*postvote 0.22*** 0.03 0.05** 0.02 — — 
Probed postvote trends       
 Linear trend −0.34** 0.12 −0.24** 0.08 −0.21*** 0.02 
 Quadratic trend 0.04* 0.02 0.02 0.01 — — 
Random effect       
 School meana 16.23 1.85 3.55 1.01 3.07 0.53 
 Residuala 31.71 1.49 18.73 1.66 19.11 1.52 
Fixed EffectsRace and/or EthnicityReligionGender
bSEbSEbSE
Prevote       
 Linear trend −1.04*** 0.13 −0.16 0.10 −0.11*** 0.02 
 Quadratic trend −0.18*** 0.02 −0.03* 0.01 — — 
Prevote-versus-postvote change       
 Postvote 0.54* 0.26 −0.13 0.21 −0.48*** 0.12 
 Linear trend*postvote 0.70*** 0.17 −0.08 0.13 −0.10** 0.03 
 Quadratic trend*postvote 0.22*** 0.03 0.05** 0.02 — — 
Probed postvote trends       
 Linear trend −0.34** 0.12 −0.24** 0.08 −0.21*** 0.02 
 Quadratic trend 0.04* 0.02 0.02 0.01 — — 
Random effect       
 School meana 16.23 1.85 3.55 1.01 3.07 0.53 
 Residuala 31.71 1.49 18.73 1.66 19.11 1.52 

Variance explained for each of the 3 types of bias-based bullying is as follows: race and/or ethnicity (0.021), religion (0.008), and gender (0.037). Analyses were restricted to students who reported being bullied because of their race and/or ethnicity, religion, or gender but not because of their sexual orientation (eg, students who reported being bullied because of their race and/or ethnicity were excluded from analyses if they also reported being bullied because of their sexual orientation). Categories are not mutually exclusive; that is, students could report multiple types of bullying as long as they did not report homophobic bullying (eg, students in the race and/or ethnicity category could also have reported being bullied because of their gender or religion). Rates of nonhomophobic bullying were estimated on the basis of the aggregated school-level data (n = 5121). Time was centered on the 2008–2009 academic year (postvote = 0 [2001–2002 through 2007–2008] versus 1 [2008–2009 through 2014–2015]). The coefficient for postvote indicates the prevote-versus-postvote intercept change. The coefficient for linear trend*postvote indicates the prevote-versus-postvote change in the linear trend; the coefficient for quadratic trend*postvote indicates the prevote-versus-postvote change in the quadratic trend. Postvote linear and quadratic trends were probed via post hoc analyses of significant interactions. The quadratic trend for bias-based bullying due to gender was also tested but was not significant and thus dropped (indicated by —).

a

Values refer to variance.

*

P < .05.

**

P < .01.

***

P < .001.

Using a quasi-experimental design with nearly 5 million youth from >5121 schools surveyed across a 14-year period, we found that the 2008–2009 academic year (during which Proposition 8 and associated activities occurred) served as a turning point for rates of homophobic bullying. We observed a 3.2–percentage-point increase (29.6% relative increase) in the proportion of students reporting homophobic bullying during the academic year that included Proposition 8 compared with 2001–2002, representing a small but robust effect size at the population level. This increase was followed by a gradual decline in homophobic bullying in the post–Proposition 8 years. Moreover, 2 sets of specificity analyses strengthened our inferences about the relationship between Proposition 8 and homophobic bullying. First, the association between Proposition 8 and bias-based bullying was not present among respondents who reported being bullied because of their race and/or ethnicity, religion, or gender but not their sexual orientation. Second, the presence of school GSAs (a factor specific to positive school climates for LGBT youth) buffered the association between Proposition 8 and homophobic bullying.

This study has several limitations. Neither the CHKS nor the GSA census data are representative samples, which may limit generalizability. However, a recent study using the representative subsample of the CHKS (which was not available during our study years) found that the prevalence of victimization is nearly identical in the representative subsample compared with the full CHKS,23 providing some evidence that our results are unlikely to be biased by potential selection of schools into the study. Furthermore, although our data comprised 14 waves of repeated cross-sectional surveys, only a prospective cohort design will allow future researchers to determine the temporal influence of contextual factors on homophobic bullying while also accounting for within-person influences.

Additionally, as in any quasi-experimental study, there is the potential that historical changes occurring alongside the Proposition 8 campaign contributed to increases in homophobic bullying and therefore serve as confounders. Although the use of ITS to examine changes before and after Proposition 8 voting is appropriate,20 the approach is stronger if there is a suitable counterfactual for comparison. In this study, the rate of homophobic bullying in secondary schools in a state that did not hold an equivalent referendum on same-sex marriage might provide such a counterfactual. Unfortunately, to our knowledge, such data do not exist. We note, however, that the Proposition 8 voter referendum in California was highly publicized and contentious, garnering national news and attention. Given the national exposure to arguments (both pro and con) regarding California’s Proposition 8, it is possible that the “best case” counterfactual does not exist even if comparable data from another state were available. Thus, future studies are needed to replicate these results and rule out plausible alternative explanations.

Finally, our data structure did not enable us to address several important questions regarding associations between Proposition 8 and homophobic bullying. These include the ability to tease apart the potentially separate effects of Proposition 8 from the campaign discourse surrounding it, to examine the immediate consequences of the repeal of Proposition 8 on homophobic bullying, and to evaluate whether Proposition 8 contributes to more pronounced elevations in homophobic bullying among sexual minority youth than their heterosexual peers. Future research with measures of sexual identity and more fine-grained data on the date of survey assessment is needed to address these questions.

This study also has a number of important strengths. It is, of course, unethical to randomly assign youth to certain social conditions, such as stigmatizing referendum and campaigns, and so randomized experiments are not possible for testing our research question. We therefore took advantage of a naturally occurring event (Proposition 8) to conduct a quasi experiment, which represents the strongest possible design to assess associations between a voter referendum that promotes stigma and homophobic bullying among youth. Quasi experiments have several advantages, particularly the ability to minimize the threat to internal validity that occurs if there is self-selection into exposure (ie, Proposition 8 referendum). We also had several years of data before and after Proposition 8 to ensure that the results were not biased by idiosyncrasies that might be apparent for the year in which the campaign occurred. An additional strength includes the large and diverse sample, which comprised nearly 5 million youth from 5121 schools.

Voter referenda, and the public campaigns that exist alongside them, occur in over half of US states. Although such referenda have frequently targeted LGBT populations,24,25 their reach is not limited to this group. In recent years, policies and campaigns related to immigration,26 criminal justice and the lives of African Americans,27 terrorism and Islamophobia,28 and bathrooms for transgender individuals29 were prominent in the public discourse. Although further research is needed to identify contextual determinants of bias-based bullying related to each of these stigmatized statuses, our results provide evidence that public campaigns communicating stigma against particular groups may confer risk for bias-based bullying among youth and suggest that the public health consequences of such frequent campaigns on common forms of peer aggression may be more wide ranging than previously realized.

Drs Russell and Hatzenbuehler conceptualized and designed the study; Drs Shen and Vandewater conducted statistical analyses; and each author drafted sections of the manuscript and approved the final manuscript as submitted.

FUNDING: The California Healthy Kids Survey was developed by WestEd under contract to the California Department of Education. The authors acknowledge generous support from the Communities for Just Schools Fund Project at the New Venture Fund and the Priscilla Pond Flawn Endowment at The University of Texas at Austin. Support for this research also was provided by grants (R24HD042849 and P2CHD042849) awarded to the Population Research Center at The University of Texas at Austin by the Eunice Kennedy Shriver National Institute of Child Health and Human Development. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Funded by the National Institutes of Health (NIH).

     
  • CHKS

    California Healthy Kids Survey

  •  
  • GSA

    gay-straight alliance

  •  
  • ITS

    interrupted time series analysis

  •  
  • LGBT

    lesbian, gay, bisexual, and transgender

1
Finkelhor
D
,
Turner
HA
,
Shattuck
A
,
Hamby
SL
.
Violence, crime, and abuse exposure in a national sample of children and youth: an update.
JAMA Pediatr
.
2013
;
167
(
7
):
614
621
[PubMed]
2
Kann
L
,
Kinchen
S
,
Shanklin
SL
, et al;
Centers for Disease Control and Prevention (CDC)
.
Youth risk behavior surveillance–United States, 2013.
MMWR Suppl
.
2014
;
63
(
4
):
1
168
[PubMed]
3
Arseneault
L
,
Bowes
L
,
Shakoor
S
.
Bullying victimization in youths and mental health problems: ‘much ado about nothing’?
Psychol Med
.
2010
;
40
(
5
):
717
729
[PubMed]
4
Lereya
ST
,
Copeland
WE
,
Costello
EJ
,
Wolke
D
.
Adult mental health consequences of peer bullying and maltreatment in childhood: two cohorts in two countries.
Lancet Psychiatry
.
2015
;
2
(
6
):
524
531
[PubMed]
5
Russell
ST
,
Sinclair
KO
,
Poteat
VP
,
Koenig
BW
.
Adolescent health and harassment based on discriminatory bias.
Am J Public Health
.
2012
;
102
(
3
):
493
495
[PubMed]
6
Swearer
SM
,
Turner
RK
,
Givens
JE
,
Pollack
WS
.
“You’re so gay!”: do different forms of bullying matter for adolescent males?
School Psych Rev
.
2008
;
37
(
2
):
160
173
7
Poteat
VP
,
Mereish
EH
,
DiGiovanni
CD
,
Koenig
BW
.
The effects of general and homophobic victimization on adolescents’ psychosocial and educational concerns: the importance of intersecting identities and parent support.
J Couns Psychol
.
2011
;
58
(
4
):
597
609
[PubMed]
8
Huang
FL
,
Cornell
DG
.
School teasing and bullying after the presidential election.
Educ Res
.
2019
;
48
(
2
):
69
83
9
Murray
M
.
Marriage rights and parental rights: parents, the state, and Proposition 8.
Stanford J Civ Rights Civ Liberties
.
2009
;
5
:
357
408
10
Staff Writer
. Timeline: Proposition 8. Los Angeles Times. August 4, 2010. Available at: http://articles.latimes.com/2010/aug/04/news/la-prop8-timeline-reverse-chron. Accessed June 23, 2018
11
Maisel
NC
,
Fingerhut
AW
.
California’s ban on same-sex marriage: the campaign and its effects on gay, lesbian, and bisexual individuals.
J Soc Issues
.
2011
;
67
(
2
):
242
263
12
Riggle
EDB
,
Rostosky
SS
,
Horne
SG
.
Marriage amendments and lesbian, gay, and bisexual individuals in the 2006 election.
Sex Res Social Policy
.
2009
;
6
(
1
):
80
89
13
Rostosky
SS
,
Riggle
ED
,
Horne
SG
,
Denton
FN
,
Huellemeier
JD
.
Lesbian, gay, and bisexual individuals’ psychological reactions to amendments denying access to civil marriage.
Am J Orthopsychiatry
.
2010
;
80
(
3
):
302
310
[PubMed]
14
Russell
GM
,
Richards
JA
.
Stressor and resilience factors for lesbians, gay men, and bisexuals confronting antigay politics.
Am J Community Psychol
.
2003
;
31
(
3–4
):
313
328
[PubMed]
15
Lipsitch
M
,
Tchetgen Tchetgen
E
,
Cohen
T
.
Negative controls: a tool for detecting confounding and bias in observational studies.
Epidemiology
.
2010
;
21
(
3
):
383
388
[PubMed]
16
Goodenow
C
,
Szalacha
L
,
Westheimer
K
.
School support groups, other school factors, and the safety of sexual minority adolescents.
Psychol Sch
.
2006
;
43
(
5
):
573
589
17
Poteat
PV
,
Sinclair
KO
,
DiGiovanni
CD
,
Koenig
BW
,
Russell
ST
.
Gay–straight alliances are associated with student health: a multischool comparison of LGBTQ and heterosexual youth.
J Res Adolesc
.
2013
;
23
(
2
):
319
330
18
Austin
G
,
Polik
J
,
Hanson
T
,
Zheng
C
.
School Climate, Substance Use, and Student Well-Being in California, 2013-2015. Results of the Fifteenth Biennial Statewide Student Survey, Grades 7, 9, and 11
.
San Francisco, CA
:
WestEd Health and Human Development Program
;
2016
19
Lopez Bernal
J
,
Cummins
S
,
Gasparrini
A
.
Interrupted time series regression for the evaluation of public health interventions: a tutorial.
Int J Epidemiol
.
2017
;
46
(
1
):
348
355
20
Penfold
RB
,
Zhang
F
.
Use of interrupted time series analysis in evaluating health care quality improvements.
Acad Pediatr
.
2013
;
13
(
suppl 6
):
S38
S44
[PubMed]
21
Zhang
F
,
Wagner
AK
,
Ross-Degnan
D
.
Simulation-based power calculation for designing interrupted time series analyses of health policy interventions.
J Clin Epidemiol
.
2011
;
64
(
11
):
1252
1261
[PubMed]
22
Stata Corp
.
Stata Statistical Software: Release 14
. 2015 ed.
College Station, TX
:
Stata Corp LP
;
2015
23
Perez-Brumer
A
,
Day
JK
,
Russell
ST
,
Hatzenbuehler
ML
.
Prevalence and correlates of suicidal ideation among transgender youth in California: findings from a representative, population-based sample of high school students.
J Am Acad Child Adolesc Psychiatry
.
2017
;
56
(
9
):
739
746
[PubMed]
24
Lewis
DC
.
Direct democracy and minority rights: same-sex marriage bans in the U.S.
Soc Sci Q
.
2011
;
92
(
2
):
364
383
[PubMed]
25
Stone
AL
.
Gay Rights at the Ballot Box
.
Minneapolis, MN
:
University of Minnesota Press
;
2012
26
Park
H
,
Parlapiano
A
. Supreme Court’s decision on immigration case affects millions of unauthorized immigrants. New York Times. June 23, 2016. Available at: www.nytimes.com/interactive/2016/06/22/us/who-is-affected-by-supreme-court-decision-on-immigration.html. Accessed June 22, 2016
27
Giriharadas
A
. Turning the call for racial reckonings back on the U.S. New York Times. July 18, 2016. Available at: www.nytimes.com/2016/07/19/us/truth-reconciliation-commission-slavery.html. Accessed July 19, 2016
28
Rappeport
A
. Ben Carson says Muslim civil rights group has ties to terrorists. New York Times. December 15, 2015. Available at: www.nytimes.com/politics/first-draft/2015/12/15/ben-carson-says-muslim-civil-rights-group-has-ties-to-terrorists/. Accessed December 15, 2015
29
The Associated Press
. Texas district denies bathroom request for transgender child. Available at: https://ktxs.com/archive/texas-district-denies-bathroom-request-for-transgender-child_. Accessed March 31, 2018

Competing Interests

POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.

FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.

Supplementary data