We used five cohorts of the Centers for Disease Control and Prevention’s (CDC’s) Youth Risk Behavior Surveillance (YRBS) national cross-sectional data from 2011 to 2019. These survey data were collected biennially, yielding a total sample size of 73 074 across these years.

The YRBS questionnaire features two items on bullying (dichotomous yes/no). It examined offline bullying with the question, “During the past 12 months, have you ever been bullied on school property?” The survey assessed online bullying with the question, “During the past 12 months, have you ever been electronically bullied? (Count being bullied through texting, Instagram, Facebook, or other social media.)”. While bullying can take a variety of forms both online and offline, including physical violence, verbal insults, harmful messages, and other forms of repeated interpersonal harm, the definition of what constitutes bullying was left to the individual respondent in the YRBS. For our analyses, being bullied online was the main predictor, while being bullied offline was one of several distinct outcomes.

Aside from offline bullying, other study outcomes related to seven types of offline experiences of injury and violence. Three of these experiences were related to weapon carrying/preparation for violence: (1) weapon carrying, “During the past 30 days, on how many days did you carry a weapon such as a gun, knife, or club?”; (2) weapon carrying at school, “During the past 30 days, on how many days did you carry a weapon such as a gun, knife, or club on school property?”; (3) carrying a gun, “During the past 12 months, on how many days did you carry a gun? (Do not count the days when you carried a gun only for hunting or for a sport, such as target shooting.)”. The item that asked about carrying a gun was only available from 2017 onward. One item was related to perceived safety: (4) avoiding school due to feeling unsafe, “During the past 30 days, on how many days did you not go to school because you felt you would be unsafe at school or on your way to or from school?”. Three items were related to being targeted for violence: (5) being threatened or injured with a weapon, “During the past 12 months, how many times has someone threatened or injured you with a weapon such as a gun, knife, or club on school property?”; (6) being in a physical fight, “During the past 12 months, how many times were you in a physical fight?”; and (7) being in a physical fight at school, “During the past 12 months, how many times were you in a physical fight on school property?”

Sex was self-reported as “female” or “male,” and race/ethnicity was based on five categories with the option to choose as many as applicable, then categorized into six mutually-exclusive categories: American Indian/Alaskan Native, Asian and Pacific Islander, non-Hispanic Black or African American, Hispanic/Latino, non-Hispanic white, and non-Hispanic Multiracial.

Sexual identity was assessed from 2015 to 2019 in the national dataset, with response options of “heterosexual (straight),” “bisexual,” “gay/lesbian,” or “not sure.”

The YRBS employs a multistage cluster sample design that featured primary sampling units of counties categorized into strata based on urbanicity and racial/ethnic composition, secondary units that pinpointed specific schools, and tertiary units of specific classes.²⁸ Sample weighting accounted for grade, race/ethnicity and sex to adjust for nonresponse and oversampling of certain groups. These weights were scaled to match national population proportions, making the sample representative of all US students in public or private school, grades 9 to 12.²⁸ Participation is voluntary and anonymous. The student response rate ranged from 80.3% (2019) to 88% (2013), while the school response rate ranged from 69% (2015) to 81% (2011). The YRBS study protocol was approved by the CDC’s Institutional Review Board.²⁸ Details on the full YRBS methodology are detailed elsewhere.²⁸

All statistical analyses accounted for the complex sampling design of the YRBS. We used the SURVEYFREQ and SURVEYLOGISTIC procedures (SAS 9.4) to incorporate sample weights and the complex sample design, estimate prevalence, and fit logistic and multinomial logistic regression models. We used the DOMAIN statement for subpopulation inference.

First, we examined the weighted prevalence of being bullied online and offline, and the other injury and violence-related behaviors, overall and by sex, race/ethnicity, and sexual identity (from 2015 to 2019). Then, we examined the association between our predictor, being a target for online bullying, and the eight outcomes (offline bullying and the seven injury/violence-related behaviors) overall, with each outcome dichotomized into “any” vs “none” using logistic regression, as well as with original response categories using multinomial logistic regression. For the multinomial models, full response categories for the offline bullying and other offline injury/violence behavior outcomes are in Supplemental Table S1. We also examined these overall associations adjusting for sex, race/ethnicity, year (2011, 2013, 2015, 2017, 2019), and grade (9th, 10th, 11th, 12th). We further examined multiplicative interactions between online bullying and demographic factors of sex, race/ethnicity, and sexual identity with each outcome and estimated the associations between online bullying and other outcomes stratified by sex, race/ethnicity, and sexual identity. It should be noted that these analyses only reflect the experiences of those who report being targeted by online bullying, lacking data on the actions of perpetrators.

Bullying and violence-related outcomes ranged in prevalence from carrying a weapon at school in the past 30 days (4.3%) to being in a physical fight within the past year (25.3%), and online bullying was associated with elevated prevalence of all outcomes. Full information on outcome prevalence, overall and by online bullying status, is available in Table 1.

Nearly two thirds (63.3%) of adolescents who reported online bullying also reported offline bullying (Figure 1). Prevalence of each violence-related outcome was also elevated; these outcomes were often more than twice as prevalent among youth experiencing bullying online.

Table 2 quantifies the associations between being bullied online and being bullied offline, weapon carrying, nonattendance due to lack of perceived safety, and being targeted for violence. Adolescents who were bullied online had 12.95 times the odds of being bullied offline compared to their peers (Table 2, 95% confidence interval [CI]: 12.03, 13.95). The associations between online bullying and the other injury and violence-related behaviors ranged from carrying a weapon overall (odds ratio [OR] = 1.35, 95% CI: 1.25, 1.46) to the violence outcome of being injured or threatened with a weapon at school (OR = 4.38, 95% CI: 3.99, 4.80). These associations remained strong, and some were stronger, after adjusting for sex, race/ethnicity, year and grade.

Supplemental Table S1 examines the association between online bullying and the number of occurrences of weapon carrying, perceived lack of safety, and being targeted for violence. For some outcomes, particularly those related to weapon carrying, the association between online bullying and each outcome level of frequency was relatively consistent. In contrast, for perceived safety and violence outcomes, the relationship was stronger for higher levels of the outcome, even after adjustment for sex, race/ethnicity, year, and grade (Supplemental Table S1, aOR [1 times being in a physical fight at school, past year vs 0]: 2.46, 95% CI: 2.18, 2.78; aOR [12+ times vs 0]: 10.39, 95%: 7.42, 14.55). Unadjusted results followed similar patterns (Supplemental Table S2).

Table 3 shows the association between each demographic factor (sex, race/ethnicity, and sexual identity) and each outcome, using “female,” “API,” and “heterosexual” as the respective reference groups given that most outcomes were least prevalent in these groups. Overall, disparities in bullying are evident and concentrated among some of the least advantaged groups. Bullying prevalence was elevated for female students, non-Hispanic Multiracial students, and sexual minority students. The violence/injury outcomes had varied associations with demographics. Generally, these outcomes were more prevalent among male students (with the exception of nonattendance due to perceived safety), American Indian and Alaskan Native students (who also had high levels of being bullied) and sexual minority students (especially for items relating to being threatened and feeling unsafe).

Table 4 summarizes the associations between online bullying and each of the other outcomes stratified by sex, race/ethnicity, and sexual identity. Interaction between sex and online bullying was significant when predicting two outcomes, the perceived safety outcome of avoiding school due to feeling unsafe (interaction p-value: .01) and physical fighting at school (interaction p-value: .02). The prevalence of each outcome by sex and online bullying status is depicted in Supplemental Figure S1, showing that the perceived safety outcome of avoiding school and the violence outcome of physical fighting at school were more closely linked to being bullied online for male students. For the association between being bullied online and school avoidance, boys had an OR of 4.31 (Table 4, 95% CI: 3.71, 5.01) compared to the OR for girls, 3.31 (95% CI: 2.88, 3.81). For the link between being bullied online and physically fighting at school, boys had an OR of 2.98 (95% CI: 2.66, 3.34) compared to the OR for girls, 2.43 (95% CI: 2.13, 2.78).

Interaction between race/ethnicity and online bullying was significant when predicting weapon carrying overall (interaction p-value <.0001) or at school (interaction p-value: .0002), and physical fighting in general (interaction p-value: .0036) or at school (interaction p-value: .02) The prevalence of each of these outcomes by race/ethnicity and online bullying status is depicted in Supplemental Figure S2, which demonstrates that the link between being bullied online and carrying a weapon in general or at school specifically was weaker among American Indian/Alaskan Native and white students (Table 4, OR between being bullied online and carrying a weapon at school, AI/AN: 1.37, 95% CI: 0.58, 3.25; white: 1.48, 95% CI: 1.24, 1.77) compared to other racial and ethnic groups (largest OR, Hispanic/Latino: 3.02, 95% CI: 2.36, 3.88). For items related to violence, such as being in a physical fight overall and at school specifically, the link between being bullied online and these outcomes was weaker for white and non-Hispanic Multiracial students (OR between being bullied online and fighting at school, Multiracial: 1.50, 95% CI: 1.04, 2.17; white: 2.25, 95% CI: 1.99, 2.54) compared to other racial and ethnic groups (largest OR, Asian and Pacific Islander: 3.26, 95% CI: 2.14, 4.95).

Interaction between sexual identity and online bullying was significant when predicting offline bullying (interaction p-value: .0036), carrying a gun (p: .02), carrying a weapon overall (p < .0001), being in a physical fight overall (p < .0001), and being in a physical fight at school (p: .01). The prevalence of each of these outcomes is depicted in Supplemental Figure S3, which shows that for many weapon carrying and violence outcomes, the links between online bullying and these outcomes were strongest among adolescents reporting their sexual identity as “not sure” (Table 4, OR between online bullying and physical fighting, “not sure”: 4.72, 95% CI: 3.31, 6.73; heterosexual: 1.80, 95% CI: 1.62, 2.01). This “not sure” group was typically followed by bisexual or gay/lesbian adolescents in terms of strength of association between online bullying and other injury/violence outcomes.

Among those who were not bullied online, the prevalence of these outcomes by sexual identity group shows very little difference; among those who were bullied online, however, the differences were pronounced. For instance, the prevalence of physical fighting at school for adolescents who were not bullied online ranged from 6.7% to 7.4% across sexual identity categories, showing very little variation. Among adolescents who were bullied online, however, the prevalence differences were severe, ranging from 13.6% among heterosexual adolescents to 27.1% in the “not sure” group, double the prevalence among heterosexual peers.

All interactions between online bullying and demographics predicting each outcome are summarized in Supplemental Table S3.

This study has limitations that should be considered when interpreting the results. These cross-sectional data do not have information on the temporal ordering between online bullying and offline injury/violence outcomes, thus the contribution of the present study is in identifying the strong links between the two, situating experiences of online bullying within broader patterns of being targeted for violence. Further research with a longitudinal design can help to disentangle the causal direction of these important connections. Several of the survey items (such as physical bullying) were specific to instances occurring on school property, and so instances of bullying outside of these spaces went unreported. Also, data only address the experiences of targets of online bullying, rather than the actions of perpetrators.

Additionally, items on injury and violence-related behaviors lacked context on the reasons behind certain actions (such as carrying a weapon in the pursuit of self-defense or for other reasons), thus making it difficult to extrapolate the motivation for engaging in these behaviors. These items also had different reference periods, with some reflecting the past month and some reflecting the past year. Demographic items were similarly limited, with no capacity to examine gender apart from sex, few sexual identity options (for instance, missing queer and asexual) starting only in 2015, and a limited race/ethnicity item.

While the YRBS aims to be nationally representative, certain years are missing data from individual states, limiting the representativeness of the findings. For instance, in 2019, Wyoming, Minnesota, Oregon, and Washington did not participate.²⁸ Furthermore, these results may not necessarily apply to other countries. However, existing international evidence suggests consistent links between adolescents being bullied online and offline experiences of bullying and violence.^38-41 Thus, the associations presented in this study should be investigated elsewhere to confirm the extent to which these connections potentially generalize to other regions.

Lastly, as these data were collected during the school days, these data also do not represent various subsets of adolescents, such as homeschooled students, homeless youth not attending school, or youth involved with the juvenile justice system who may not be in school.

Being bullied online is significantly associated with offline injury and violent behaviors in and out of school. Moreover, these associations are patterned by demographic factors, with online bullying predicting many of these behaviors for boys, sexual minority adolescents, and adolescents of color more strongly. Efforts to significantly reduce online bullying are an urgent public health priority with the potential to limit troubling in-person behaviors such as physical violence and weapon carrying. These goals are especially important for marginalized adolescents who may be particularly impacted by online bullying and its consequences.