Risk and Protective Factors for Bullying, Cyberbullying, and Sexual Harassment in U.S. Schools: A Socio-Ecological Analysis Beyond Individual Level Factors

Abstract

This study identifies the socio-ecological predictors at each level of the ecosystem that serve as risk or protective factors for the frequency of bullying, cyberbullying, and sexual harassment in U.S. schools, and examines how these predictors vary across the three behaviors, highlighting both common and distinct predictors. Using data from two waves (2018 and 2020) of the School Survey on Crime and Safety (SSOCS), a nationally representative cross-sectional survey of 5132 U.S. public elementary and secondary schools, we conduct a Generalized Ordinal Logistic Regression in hierarchical blocks of mesosystem-, exosystem-, and macrosystem-level factors to examine the relative contribution of each ecological level. We find that both mesosystem (teacher training, school size, school grade) and exosystem (neighborhood crime, parental involvement, and involvement of mental health & social services)-level factors play significant roles in predicting the frequency of bullying, sexual harassment, and cyberbullying. Urbanicity at the macrosystem-level is not a significant predictor of bullying and sexual harassment; however, schools in urban areas are less likely to have frequent and monthly cyberbullying and more likely to have occasional and no cyberbullying compared to schools in non-urban areas. We argue for the importance of multi-level interventions to address peer harassment in schools through evidence-based policies.

1. Introduction

Bullying, cyberbullying, and sexual harassment remain pervasive public health and educational concerns in U.S. schools (Bolduc et al., 2023; DeGue et al., 2021). These forms of peer harassment have been consistently linked to a range of adverse outcomes, including diminished academic engagement, psychological distress, and long-term mental health difficulties (Basile et al., 2020; Valik et al., 2023). Despite widespread awareness of their consequences, the underlying mechanisms that sustain these behaviors across school contexts are not fully understood. Traditional approaches have largely focused on individual-level predictors—such as student demographics, personality traits, and victimization histories—often overlooking the broader ecological factors that shape school climates and student interactions (Sahin-Ilkorkor & Brubaker, 2025).

The socio-ecological framework offers a valuable lens for examining how factors at multiple levels—ranging from interpersonal relationships to institutional and community environments—interact to influence peer harassment. Within this framework, the mesosystem encompasses features internal to the school, such as teacher training, disciplinary practices, school size, and grade level; the exosystem extends to the surrounding neighborhood, community, and parental engagement; and the macrosystem reflects broader structural characteristics, including urbanicity and regional norms. Understanding how these levels shape outcomes collectively is critical for designing effective prevention and intervention strategies that extend beyond the individual.

This study applies a socio-ecological approach to investigate and compare the determinants of bullying, cyberbullying, and sexual harassment across 5132 U.S. public elementary and secondary schools using data from two waves (2018 and 2020) of the School Survey on Crime and Safety (SSOCS), a nationally representative cross-sectional survey of U.S. public elementary and secondary schools. Our first research question asks which socio-ecological factors at each level serve as risk or protective factors for the frequency of bullying, cyberbullying and sexual harassment in schools? Our second question asks how do socio-ecological predictors vary across bullying, cyberbullying, and sexual harassment, highlighting both common and distinct predictors?

The article is organized as follows: first, we provide an overview of the conceptual framework, the socio-ecological model, as applied to educational settings. Next, we review the relevant literature on risk and protective factors for bullying, cyberbullying, and sexual harassment in schools at each of the socio-ecological levels, with an emphasis on the levels under examination, that contribute to our hypotheses. We then describe our research design, followed by an overview of the design and implementation of the SSOCS survey, and we identify the variables that we utilize for the Hierarchical Generalized Ordinal Logistic Regression model. We then share the findings of the analysis and discuss policy, practice, and research implications of this deeper understanding of risk/protective factors for these specific forms of school-based violence.

2. Conceptual Framework

Scholars from various disciplines utilize the socio-ecological model to identify risk and protective factors for a range of outcomes across the various levels of human development and interaction. Initially developed by Bronfenbrenner (1979), the socio-ecological framework views human behaviors and decision-making as occurring within a broad context of interconnected interactions and influences from the individual level to the interpersonal, community, and societal levels. Especially prominent within the field of public health, the socio-ecological model has also been used to better understand factors that contribute to a variety of experiences within the educational context.

Allen et al. (2018) have applied the socio-ecological model to the school setting by specifying the focus of each of the levels. The individual level focuses on the student and includes factors such as their emotional stability, personality characteristics, and academic motivation, and this level has received the most attention from researchers. Based on a narrative review of the literature, we reported in an earlier publication (Sahin-Ilkorkor & Brubaker, 2025) that most of the published studies on school-based bullying and harassment focus on individual-level factors such as demographic characteristics of children at greatest risk for victimization, and such studies rely primarily on students’ reports of their own subjective experiences.

The microsystem includes factors such as teacher, parental, and peer support. The mesosystem refers to school policies, extracurricular activities, staff and teacher training, and specific rules and practices. The exosystem includes broad school-level factors such as its mission/vision, the school board’s decisions/political approach, and the surrounding community. The macrosystem is even broader and can include the social and political climate and historical development of the setting as well as government policies and reforms and legislation.

Guided by Allen et al.’s (2018) interpretation of the socio-ecological model, we focus on factors across the social-ecological spectrum as enhancing risk for, or providing protection from, bullying, cyberbullying, and sexual harassment in schools. Given the lack of research on factors beyond the individual-level, and based on the specific data source for our study, we focus primarily on the meso-, exo- and macrosystem factors.

3. Literature Review

Researchers use fairly consistent definitions of these forms of behaviors. Bullying, for example, generally emphasizes repeated behaviors between individuals who are not siblings or romantic partners that are intended to cause harm to the victim and carried out in a context of power imbalance (Gladden et al., 2014). Cyberbullying definitions include the focus on “deliberate harmful behavior, carried out repeatedly, where there is a perceived or actual imbalance in power against a target who is vulnerable or cannot easily stand up to the perpetrator” and adds that this form of harm is “inflicted through the use of computers, cell phones, and other electronic devices” (Margolis & Amanbekova, 2023, p. 79). Sexual harassment, on the other hand, is specifically defined by the U.S. Department of Education as behavior that “is sexual in nature; is unwelcome; and denies or limits a student’s ability to participate in or benefit from a school’s education program” (U.S. Department of Education & Office for Civil Rights, 2008, p. 3).

The increasing prevalence of cyberbullying adds to the definitional and practical challenges to understanding the risk and protective factors of these behaviors. Although similar to traditional bullying, cyberbullying adds protections of anonymity that embolden perpetrators and expand the scope of harm beyond physical school spaces and outside of the purview of teachers and other guardians (Margolis & Amanbekova, 2023; Nagata et al., 2022). In fact, cyberbullying has been found to be more stressful and to cause more serious consequences compared to traditional bullying (Guo et al., 2021).

More recently, UNESCO has provided a more inclusive and comprehensive definition of bullying:

“School bullying is a damaging social process that is characterized by an imbalance of power driven by social (societal) and institutional norms. It is often repeated and manifests as unwanted interpersonal behaviour among students or school personnel that causes physical, social, and emotional harm to the targeted individuals or groups, and the wider school community.”

(UNESCO, 2023, para. 12)

The authors suggest that this “proposed definition promotes a holistic and inclusion-driven approach to tackling bullying and violence in schools and in online spaces” (UNESCO, 2023, para. 6).

In this study, we rely on the SSOCS survey’s definitions of these behaviors and the respondents’ reports of the types of behaviors occurring in their schools. In the following sections, we review literature on these forms of school-based violence that have identified risk and protective factors at each of the social-ecological levels, with the exception of the individual level since our dataset does not include individual-level data. We give particular attention to the school- and community-based factors—as they are those most addressed in the survey and the factors that have received the least attention from researchers.

3.1. Microsystem—Student Interactions with Parents, Teachers and Peers

A few researchers have examined the influence of teachers, peers, and parents on bullying and sexual harassment experiences, but the focus has been more on the nature of students’ relationships with others than on their direct influence. For example, studies have found that students who witness family violence are more likely to engage in all forms of school-based violence as perpetrators and as victims (Baldry, 2003; Cook et al., 2010; Fineran & Bolen, 2006; Lereya et al., 2013; Margolis & Amanbekova, 2023). Researchers have also found that positive relationships and open communication with parents can protect children from bullying (Doty et al., 2017; Guo et al., 2021; Khairi et al., 2022; Lereya et al., 2013). Others have found that parents’ oversight and enforcement of rules can be a deterrent from perpetrating bullying (Hinduja & Patchin, 2013; Schilling & Wang, 2023). Some research connects gender socialization by parents to experiences with sexual harassment (Brown et al., 2020).

In terms of teachers’ interactions, research has suggested teachers’ clear communication prohibiting bullying behavior and swift responses when bullying occurs can serve as protective factors against bullying (Harasgama & Jayathilaka, 2023; Khairi et al., 2022; Olweus, 1993). Students who report that their teachers are fair and supportive, and that they feel their teachers care about them are also less likely to engage in bullying (Barboza et al., 2009; Doty et al., 2017). The literature on staff members’ teachers’ influence on sexual harassment suggests that the likelihood of these behaviors increases when teachers dismiss, ignore, or tolerate them (DeSouza & Ribeiro, 2005; Horn & Poteat, 2023; Ormerod et al., 2023). Some researchers suggest that teachers themselves feel ill-prepared to identify or respond to sexual harassment between students and tend to believe that it happens more among adults (Brown et al., 2020; Charmaraman et al., 2013; Edwards et al., 2020). Studies do suggest, however, that similarly to bullying, when students believe that teachers and other staff take sexual harassment seriously and respond with punitive measures, the prevalence of these behaviors is lower (Skoog et al., 2023).

Student interactions with peers can also serve as protective or risk factors for these forms of violence, depending on the nature and quality of the relationships. Prosocial, positive and supportive relationships with peers can serve as protective factors (Ttofi et al., 2014), where relationships with anti-social peers who condone and encourage bullying and sexual harassment increase the risk for engaging in these behaviors (Bollmer et al., 2005; Hinduja & Patchin, 2013; Kendrick et al., 2012).

The SSOCS does not include many questions that address the micro level influences on violent behaviors in schools. One exception is that the survey includes two questions about parental influences: one focuses on formal processes that solicit parental input into policies regarding school safety, and the other on providing assistance to parents in addressing students’ problematic behaviors. These questions do not directly measure the quality of parent-student interactions. Instead, they reflect how schools formally engage parents at an institutional level. Hence, we address these parental indicators as “community involvement” and an aspect of the exosystem in our study.

3.2. Mesosystem—School Features

Some researchers have examined features of schools themselves to better understand the contexts in which these behaviors occur. For example, some studies attempt to measure the school climate, which Cohen et al. (2009) suggest “is based on patterns of people’s experiences of school life and reflects norms, goals, values, interpersonal relationships, teaching and learning practices, and organizational structures” (p. 180). This definition includes factors across the socio-ecological spectrum and can include both organizational and social dimensions, and research suggests that students who feel less positive about their schools and have a lower sense of belonging are a greater risk for bullying perpetration and victimization (Hong & Espelage, 2012; Barboza et al., 2009; O’Brien, 2021; Cook et al., 2010). Other mesosystem factors include the school’s prevention policies related to these behaviors and the training provided to teachers and other staff to identify and respond to bullying and harassment (Gruber & Fineran, 2016; Brown et al., 2023). States vary in their anti-bullying laws and bullying definitions, the procedures they use for reporting and investigation procedures, and the provisions governing prevention and training (Cascardi et al., 2018). Some researchers have studied schools’ compliance with these policies, as well as teachers’ and students’ awareness of such policies (Brown et al., 2023).

The SSOCS identifies several programs that schools might employ to address various forms of violence including specific prevention curricula focused on forms of violence, social and emotional learning programs, student courts, restorative practices, and “programs to promote a sense of community or social integration among students” (Kaatz et al., 2024).

3.3. Exosystem—Community/Neighborhood

There is limited research on this level of influence on school-based violence, but this level can include a broad range of factors focused on community involvement in promoting school safety, as well as measures of neighborhood crime in the area where the school is located. “Community” encompasses a variety of entities, including youth-supportive organizations and faith-based organizations, as well as local law enforcement agencies.

Some research suggests that witnessing neighborhood violence, higher levels of neighborhood disorder, and lack of adult supervision, and exposure to older and larger adolescents are risk factors for experiencing bullying (Chaux et al., 2009; Davis et al., 2020; D’Urso et al., 2021; Schilling & Wang, 2023; Schwartz et al., 2021). Lower socioeconomic levels and lower levels of social capital have been found to be risk factors as well (D’Urso et al., 2021; Han et al., 2019; Nagata et al., 2022). Although little research has focused on community level protective factors, Johns et al. (2020) suggest that schools’ partnerships with local organizations providing support to marginalized youth can serve as protective factors.

The SSOCS addresses the involvement of various groups in promoting school safety, including parent groups, social service agencies, juvenile justice agencies, law enforcement agencies, mental health agencies, civic organizations/service clubs, private corporations or businesses, and religious organizations. In our study, we distinguish between groups providing supportive resources and services and those focused on law enforcement and punishment.

3.4. Macrosystem—Broader Society

Some researchers have examined broader social and cultural factors that contribute to bullying and sexual harassment, such as gender and sexual norms and mass media that promote traditional masculinity and heteronormativity. For example, normative beliefs about bullying, as well as broader cultural beliefs regarding sexual and gender identities have been found to contribute to harassing behaviors (Hong & Garbarino, 2012; Phoenix et al., 2003; K. R. Williams & Guerra, 2007). D’Urso et al. (2021) also suggest that conflicting cultural norms between immigrant children’s culture of origin and those of the culture of settlement can contribute to bullying experiences. Huang and Cornell (2019) studied the influence of political attitudes on bullying and found that there were higher rates of bullying incidents in counties supporting the Republican candidate after the 2016 U.S. presidential election. Similarly, Herrera Hernandez and Oswald (2023) found relationships between young people’s attitudes towards the confirmation of U.S. Supreme Court Justice Brett Kavanaugh in 2018 and their attitudes toward sexual harassment.

A few studies have examined geographical location as an influence on bullying, yielding mixed findings (Cabrera et al., 2024). Early studies focused on cyberbullying have speculated that this form of violence would be more prevalent based on the greater access to the internet in urban areas, but these disparities have largely disappeared in most developed countries. A more recent study in Spain focused on middle and high school students (Cabrera et al., 2024) found that more students in urban than rural areas reported bullying perpetration, cyberbullying was more often directed at peers in rural areas, and the impact of cyberbullying caused more distress among students in rural areas.

The SSOCS does not include variables related to these broader social and cultural factors, with the exception of geographic context, i.e., urban v. rural. Building on research on bullying and cyberbullying, we extend our examination to sexual harassment and look for variation of all three forms of harassment by school location. In addition, we propose to use this variable as a proxy for political climate given the general association between greater diversity in terms of race, gender, and sexual identity, and greater support for inclusive policies in urban areas (Huijsmans & Rodden, 2025).

4. Methods and Design

4.1. Data

The present study utilizes cross-sectional data from the School Survey on Crime and Safety (SSOCS), a nationally representative survey of K–12 public schools in the United States. The unit of analysis is the school. This dataset provides information about violent crimes at schools, including bullying and peer harassment, characteristics of the school environment, and policies and practices employed by schools to prevent and reduce school crime (Padgett et al., 2020; Kaatz et al., 2024). The SSOCS survey itself frames these behaviors in specific ways throughout its sections and questions, further discussed in the dependent variable section below. It provides a list of specific kinds of behaviors that fall under the broad umbrella of bullying and harassment and asks respondents to provide their perception of how frequently each type occurs at their school. The assumption behind the survey design is that these behaviors are distinct and that the survey respondent can clearly identify each type of behavior. While respondents are asked to count the number of formal incidents that are reported that constitute crimes, they are asked to provide their perception of the frequency of these behaviors which do not appear to be based on official reports.

The SSOCS survey has been conducted in odd years since 1999 and the latest dataset, which has become publicly available since November 2024, covers the school year of 2019–20 (Kaatz et al., 2024). SSOCS is administered by the United States Census Bureau and sponsored by the National Center for Education Statistics (NCES). Public schools are randomly selected for participation by the U.S. Census Bureau using stratified random sampling. Stratification is performed based on school level (primary, middle, high, and combined schools), locale (city, suburb, town, and rural), and enrollment size (Padgett et al., 2020; Kaatz et al., 2024). After being included in the sample, all schools are surveyed via mailed paper questionnaire and online questionnaire (Padgett et al., 2020; Kaatz et al., 2024).

Although the 2019–20 SSOCS was administered during the onset of the COVID-19 pandemic, most of the survey responses cover the period before the school closures and provide information about conditions during a typical school year. Most schools completed the survey in February (34%) and March (25%) before or in the very early stages of the pandemic, followed by April (7%) and May (7%) (Kaatz et al., 2024). Additionally, the National Center for Education Statistics performed some reviews and analyses to address concerns about the anomalies associated with the pandemic. Their analysis did not find any problem with data quality and characteristics of school respondents before and after the pandemic (Kaatz et al., 2024). Their benchmarking analysis also did not find any problem with the trends seen in the data across years (Kaatz et al., 2024). The NCES notes that “these reviews did not unearth any indicators of data quality issues related to the pandemic that would require any adjustments to the data” (Kaatz et al., 2024, p. 44). Therefore, we included the 2019–20 dataset to ensure our analysis reflects the most recent available data, which is especially relevant given the growing importance of cyberbullying.

The present study uses the latest publicly available SSOCS datasets and combines the datasets covering the 2017–18 school year, and 2019–20 school year. The total sample size of our study is 5132 schools. 2762 schools out of a stratified, random sample of 4803 schools completed the survey in 2018 (response rate of 61.7%) (Padgett et al., 2020). 2370 schools out of a stratified, random sample of 4800 schools completed the survey in 2020 (response rate of 54.1%, that was somewhat lower than the response rate of 2018 due to pandemic-related disruptions) (Kaatz et al., 2024). Among the 5132 schools, approximately 25% were elementary schools, 35% middle schools, 36% high/secondary schools, and 4% combined/other schools. These schools were also stratified by enrollment size (11% enrolling fewer than 300 students, 22% enrolling 300–499 students, 39% enrolling 500–999 students, and 28% enrolling 1000 or more) and geographic spread (26% of schools were located in cities, 37% in suburbs, 14% in towns, and 23% in rural areas). The distribution of schools by grade level, size, and geographic locale is highly consistent across the 2017–18 and 2019–20 survey waves. Datasets remain representative of U.S. public schools because the sample design and weighting adjustments account for nonresponse due to pandemic-related disruptions (Kaatz et al., 2024). For transparency, descriptive statistics for the 2017–18 and 2019–20 samples are presented separately in the Appendix A,

Table A1. Descriptive Statistics of Study Variables (Wave 2018 vs. Wave 2020).

	Wave 2018 (N = 2762)		Wave 2020 (N = 2370)
	Frequency	Percent	Frequency	Percent
Dependent Variables
Bullying Frequency
(0 = never)	83	3.01%	53	2.24%
(1 = occasional)	1592	57.64%	1453	61.31%
(2 = monthly)	559	20.24%	424	17.89%
(3 = frequent)	528	19.12%	440	18.57%
Sexual Harassment Frequency
(0 = never)	824	29.83%	694	29.28%
(1 = occasional)	1683	60.93%	1467	61.90%
(2 = monthly)	189	6.84%	139	5.86%
(3 = frequent)	66	2.39%	70	2.95%
Cyberbullying Frequency
(0 = never)	218	7.89%	124	5.23%
(1 = occasional)	1196	43.30%	1089	45.95%
(2 = monthly)	638	23.10%	546	23.04%
(3 = frequent)	710	25.71%	611	25.78%
Mesosystem-Level Factors
Teacher Training Score	Mean 9.36; Std. dev. 3.31		Mean 10.20; Std. dev. 3.04
School Safety & Climate Score	Mean 6.64; Std. dev. 2.06		Mean 7.25; Std. dev. 2.00
School Disciplinary Practices
Security cameras (1 = Yes)	2464	89.21%	2217	93.54%
Threat reporting system (1 = Yes)	1573	56.95%	1744	73.59%
Threat assessment team (1 = Yes)	1349	48.84%	1622	68.44%
Sworn law enforcement officers (1 = Yes)	1859	67.31%	1614	68.10%
Metal detector checks (1 = Yes)	233	8.44%	246	10.38%
Grade Level of School
(1 = elementary)	671	24.29%	624	26.33%
(2 = middle)	975	35.30%	818	34.51%
(3 = high/secondary school)	997	36.10%	827	34.89%
(4 = combined/other)	119	4.31%	101	4.26%
School Size
(1 = school size less than 300)	286	10.35%	262	11.05%
(2 = school size of 300–499)	605	21.90%	507	21.39%
(3 = school size of 500–999)	1042	37.73%	974	41.10%
(4 = school size more than 1000)	829	30.01%	627	26.46%
Exosystem-Level Factors
Parental Involvement
(0 = none)	441	15.97%	273	11.52%
(1 = moderate)	1651	59.78%	1325	55.91%
(2 = high)	670	24.26%	772	32.57%
Community Involvement
Juvenile justice & law enforcement (1 = Yes)	2371	85.84%	2139	90.25%
Mental health & social services (1 = Yes)	2215	80.20%	2118	89.37%
Civic & businesses & religious org. (1 = Yes)	1662	60.17%	1516	63.97%
Neighborhood Crime
(1 = low)	2035	73.68%	1758	74.18%
(2 = moderate)	569	20.60%	450	18.99%
(3 = high)	158	5.72%	162	6.84%
Macrosystem-Level Factors
Urbanicity (1 = Yes)	1757	63.61%	1497	63.16%
City	723	26.18%	633	26.71%
Suburb	1034	37.44%	864	36.46%
Urbanicity (0 = No)	1005	36.39%	873	36.84%
Town	382	13.83%	325	13.71%
Rural	623	22.56%	548	23.12%

Note 1: The two survey waves are highly comparable in terms of overall school characteristics. The largest differences are observed in the presence of threat reporting systems and threat assessment teams: 57% and 49% of schools reported having these measures in 2018, compared to 74% and 68% in 2020, respectively.

.

4.2. Study Variables

4.2.1. Dependent Variables

Frequency of Incidents: We have three dependent variables: the frequency of bullying, the frequency of sexual harassment, and the frequency of cyberbullying. Definitions of these incidents are provided at the beginning of the SSOCS survey. Bullying is defined as “any unwanted, aggressive behavior(s) by another youth or group of youths that involves an observed or perceived power imbalance and is repeated multiple times or is highly likely to be repeated. Bullying occurs among youth who are not siblings or current dating partners” (Kaatz et al., 2024, p. A-3). Cyberbullying is defined as “bullying that occurs when willful and repeated harm is inflicted through the use of computers, cell phones, or other electronic devices” (Kaatz et al., 2024, p. A-3). Sexual harassment is defined as “conduct that is unwelcome, sexual in nature, and denies or limits a student’s ability to participate in or benefit from a school’s education program. All students, regardless of sex or gender identity, can be victims of sexual harassment, and the harasser and the victim can be of the same sex. The conduct can be verbal, non-verbal, or physical and can take many forms, including verbal acts and name-calling, as well as nonverbal conduct, such as graphic and written statements, or conduct that is physically threatening, harmful, or humiliating.” (Kaatz et al., 2024, p. A-4).

Each variable was measured on four ordered categories (0 = Never, 1 = occasional, 2 = monthly, 3 = frequent). These variables are based on survey items that ask schools how often these incidents occur at their schools. The response options of these survey items used a 5-point Likert scale, which collapsed into four categories to have enough cases in each category while acknowledging the possible loss of information. Specifically, response options of “happens daily” and “happens at least once a week” were recoded as “frequent”, while “happens at least once a month” was coded as “monthly”, “happens on occasion” was coded as “occasional”, and “never happens” was coded as “never”.

4.2.2. Independent Variables

Mesosystem-Level Factors

Teacher Training: This is a composite index score measured as the sum of 13 survey items whether the school or school district provides any of the following training for classroom teachers or aides: training in school-wide discipline policies and practices related to (1) violence, (2) cyberbullying, and (3) bullying other than cyberbullying, (4) alcohol and/or drug use, (5) training in safety procedures, (6) training in recognizing early warning signs of students likely to exhibit violent behavior, (7) training in recognizing signs of self-harm or suicidal tendencies, (8) training in intervention and referral strategies for students displaying signs of mental health disorders (e.g., depression, mood disorders, ADHD), (9) training in recognizing physical, social, and verbal bullying behaviors, (10) training in recognizing signs of students using/abusing alcohol and/or drugs, (11) training in positive behavioral intervention strategies, (12) training in crisis prevention and intervention, and (13) training in classroom management. This composite score ranges from 0 to 13 and higher values indicate a more comprehensive teacher training about preventing, identifying, or managing violent or aggressive behaviors. This variable is constructed through factor analysis, and internal consistency reliability is calculated through Cronbach’s alpha score. Cronbach’s alpha score for this variable is 0.85, suggesting a high reliability in the present study.

School Safety & Climate: This is a composite index score measured as the sum of 11 survey items whether the school has any activities/policies for students: (1) Prevention curriculum, instruction, or training for students such as conflict resolution, anti-bullying, dating violence prevention, (2) Social emotional learning (SEL) for students such as social skills, anger management, mindfulness, (3) Behavioral or behavior modification intervention for students including use of positive reinforcements, (4) Individual mentoring, tutoring or coaching of students by adults, (5) Student involvement in peer mediation, (6) Student court to address student conduct problems or minor offenses, (7) Student involvement in restorative practices such as peace circles or conflict circles), (8) Programs to promote a sense of community or social integration among students, and recognized student groups with the acceptance of (9) sexual orientation and gender identity of students, (10) students with disabilities, and (11) cultural or religious diversity. This composite score ranges from 0 to 11, and higher values indicate a more comprehensive prevention and sense of community policy. This variable is constructed through factor analysis, and internal consistency reliability is calculated through Cronbach’s alpha score. While the internal consistency is modest (Cronbach’s alpha score = 0.651), this level is considered acceptable for multidimensional policy constructs. Factor analysis supports a one-factor solution, and we have retained all items to ensure comprehensive coverage of prevention and sense of community strategies.

School Discipline-Related Practices: Each school disciplinary practice is treated as a distinct binary indicator. Specifically, each binary variable shows whether the school implements a particular practice, including (i) using one or more security cameras to monitor the school, (ii) providing a structured anonymous threat reporting system (e.g., online submission, telephone hotline, or written submission via drop box), (iii) having a threat assessment team or any other formal group of persons to identify students who might be a potential risk for violent or harmful behavior (toward themselves or others), (iv) having any sworn law enforcement officers (including School Resource Officers) present at the school at least once a week, and (v) performing regular or random metal detector checks on students. Each school disciplinary practice is treated as a distinct indicator due to low internal consistency among these items.

Structural Features of the School Environment: School grade level (elementary, middle, high/secondary school, and combined/other) and school enrollment size (<300, 300–499, 500–999, bigger than 1000 students) are taken into account as school-level variables in the mesosystem.

Exosystem-Level Factors

Neighborhood Crime: This is a categorical variable showing crime level in the area where the school is located (1 = low level of crime, 2 = moderate level of crime, 3 = high level of crime).

Community Involvement: Community involvement is disaggregated into four subgroups based on the functional roles of these groups, with each group being a separate predictor. (i) Parental involvement is a categorical variable measured by using three survey items of whether the school (1) has a formal process to obtain parental input on policies related to school crime and discipline, (2) provides training or technical assistance to parents in dealing with students’ problem behavior, and (3) involves parents groups in the school’s efforts to promote a safe school. Parental involvement is “none” if the school does not have any of these practices, “moderate” if the school has one or two practices, and “high” if the school has all three practices. (ii) Juvenile justice & law enforcement involvement is a binary variable taking a value of 1 if the school involves juvenile justice agencies or law enforcement agencies in the school’s efforts to promote a safe school. This variable reflects an external community partnership where schools collaborate with these agencies in addressing behavioral or safety issues beyond daily operations (e.g., providing input on school crime and discipline policies). The presence of sworn law enforcement officers at schools, on the other hand, reflects internal safety practices embedded in daily school operations (e.g., routinely monitoring the school environment). Therefore, community involvement of juvenile justice and law enforcement involvement is part of exosystem-level factors, whereas the presence of sworn law enforcement officers at the school is part of mesosystem-level factors. (iii) Mental health & social services involvement is a binary variable taking a value of 1 if the school involves mental health agencies or social service agencies in the school’s efforts to promote a safe school. (iv) Civic and other community involvement is a binary variable taking a value of 1 if the school involves civic organizations or service clubs, private corporations or businesses, or religious organizations.

Macrosystem-Level Factors

Urbanicity: This is a binary variable which takes a value of 1 if the school is located in an urban area (city or suburb) and takes the value of 0 if the school is located in a non-urban area (town or rural).

Time: Year-specific shocks (e.g., national policies or social trends) are taken into account by having a binary variable for the wave of survey (dataset of 2018 vs. dataset of 2020).

4.3. Data Analysis Plan and Methodology

The study uses a non-experimental, associational, cross-sectional research design and conducts an ordinal logistic regression analysis. The 2018 and 2020 datasets were combined to have enough cases in each response category of our dependent variables. For each variable in our analysis, item response rates exceeded 99 percent, and missing values were imputed by the National Center for Education Statistics before releasing the dataset (Padgett et al., 2020; Kaatz et al., 2024). The National Center for Education Statistics adopted a direct copy of the imputation method and used data from a similar case to generate an imputed value for each missing value for our study variables (Padgett et al., 2020; Kaatz et al., 2024). Descriptive statistics and zero-order correlations for study variables were presented. Factor analysis was conducted, and Cronbach’s alpha score was calculated to construct composite index scores.

Initially, the Ordered Logit Model was applied because dependent variables (frequency of incidents) were measured on an ordinal scale (0 = Never, 1 = occasional, 2 = monthly, 3 = frequent). After testing the proportional odds assumption, a generalized ordinal logistic regression was estimated to account for a partial proportional odds model (R. Williams, 2006). The Wald tests in the final models indicated no substantial violation of the proportional odds assumption after partial constraints were applied. The Generalized Ordinal Logistic Regression was implemented in hierarchical blocks for mesosystem-, exosystem-, and macrosystem-level factors to examine the relative contribution of each ecological level. The model fit was presented using the likelihood ratio test, AIC/BIC, and McFadden’s R-Squared. Cluster standard errors at the school level were applied to account for repeated observations of the same school across years. A p-value of <0.05 was considered significant. All analyses were conducted in STATA Version 19.

As a robustness check, pandemic-related school closures were taken into account. 59% of the SSOCS 2020 survey responses were retained after excluding survey responses that were completed after school closures. The results were substantively unchanged, suggesting that pandemic-related disruptions did not bias our findings (please see the Appendix A,

Table A2. Marginal Effects of Significant Study Variables (Restricted Sample of 4153).

	Bullying Frequency		Sexual Harassment Frequency		Cyberbullying Frequency
Common Protective Factors
Teacher Training Score
never	0.001	(0.00) ***	0.004	(0.00) *	0.001	(0.00) **
occasional	0.010	(0.00) ***	−0.002	(0.00) *	0.004	(0.00) **
monthly	−0.004	(0.00) ***	−0.001	(0.00) *	−0.001	(0.00) *
frequent	−0.007	(0.00) ***	−0.001	(0.00)	−0.005	(0.00) **
High Parental Involvement (compared to no parental involvement)
never	0.004	(0.00)	0.048	(0.02) *	0.023	(0.01) ***
occasional	0.030	(0.02)	−0.025	(0.02)	0.061	(0.02) ***
monthly	−0.011	(0.01)	−0.028	(0.01) **	−0.013	(0.00) ***
frequent	−0.023	(0.02)	0.005	0.006	−0.071	0.019***
Existence of Metal Detector Checks
never	0.010	(0.00) *	0.077	(0.02) **	0.015	(0.01) *
occasional	0.064	(0.02) **	−0.047	(0.02) **	0.037	(0.02) *
monthly	−0.027	(0.01) **	−0.022	(0.01) ***	−0.009	(0.01)
frequent	−0.047	(0.02) **	−0.009	(0.00) ***	−0.042	(0.02) *
Common Risk Factors
Middle School (compared to Elementary School)
never	−0.027	(0.00) ***	−0.348	(0.02) ***	−0.168	(0.01) ***
occasional	−0.224	(0.02) ***	0.243	(0.02) ***	−0.275	(0.02) ***
monthly	0.088	(0.01) ***	0.074	(0.01) ***	0.172	(0.02) ***
frequent	0.164	(0.01) ***	0.031	(0.00) ***	0.271	(0.01) ***
High/Secondary School (compared to Elementary School)
never	−0.015	(0.00) ***	−0.366	(0.02) ***	−0.160	(0.01) ***
occasional	−0.082	(0.02) ***	0.306	(0.02) ***	−0.208	(0.02) ***
monthly	0.041	(0.01) ***	0.045	(0.01) ***	0.148	(0.02) ***
frequent	0.055	(0.01) ***	0.015	(0.00) ***	0.219	(0.02) ***
School Size more than 1000 (compared to school size less than 300)
never	−0.027	(0.01) ***	−0.227	(0.03) ***	−0.067	(0.01) ***
occasional	−0.160	(0.02) ***	0.137	(0.02) ***	−0.163	(0.02) ***
monthly	0.066	(0.01) ***	0.063	(0.01) ***	0.055	(0.01) ***
frequent	0.120	(0.02) ***	0.026	(0.00) ***	0.175	(0.02) ***
High Neighborhood Crime (compared to low neighborhood crime)
never	−0.002	(0.01)	−0.086	(0.02) ***	−0.025	(0.01) ***
occasional	−0.120	(0.03) ***	0.037	(0.01) ***	−0.074	(0.02) **
monthly	−0.042	(0.02)	0.034	(0.01) **	0.009	(0.00) ***
frequent	0.164	(0.03) ***	0.015	(0.01) **	0.090	(0.03) **
Existence of Mental Health & Social Services Involvement
never	−0.011	(0.00) ***	−0.059	(0.02) **	−0.026	0.006***
occasional	−0.074	(0.02) ***	0.035	(0.01) **	−0.063	0.013***
monthly	0.031	(0.01) ***	0.017	(0.00) ***	0.018	(0.00) ***
frequent	0.055	(0.01) ***	0.007	(0.00) ***	0.071	(0.01) ***
Distinct Protective & Risk Factors
Existence of Security Cameras
never	0.004	(0.00)	0.034	(0.02)	−0.004	(0.01)
occasional	0.032	(0.02)	−0.016	(0.01)	−0.010	(0.02)
monthly	−0.011	(0.01)	−0.012	(0.01)	0.002	(0.00)
frequent	−0.024	(0.02)	−0.005	(0.00)	0.012	(0.02)
Existence of Threat Reporting System
never	0.000	(0.00)	−0.013	(0.01)	−0.007	(0.00) *
occasional	−0.003	(0.01)	0.007	(0.01)	−0.019	(0.01) *
monthly	0.001	(0.01)	0.004	(0.00)	0.004	(0.00)
frequent	0.002	(0.01)	0.002	(0.00)	0.022	(0.01) *
Existence of Sworn Law Enforcement Officers
never	−0.018	(0.01) **	−0.001	(0.01)	−0.012	(0.00) **
occasional	0.014	(0.02)	0.000	(0.01)	−0.033	(0.01) **
monthly	−0.004	(0.01)	0.000	(0.00)	0.008	(0.00) *
frequent	0.008	(0.01)	0.000	(0.00)	0.037	(0.01) **
Existence of Juvenile Justice & Law Enforcement Involvement
never	−0.002	(0.00)	−0.021	(0.02)	−0.008	(0.01)
occasional	−0.017	(0.02)	0.011	(0.01)	−0.021	(0.01)
monthly	0.006	(0.01)	0.007	(0.01)	0.005	(0.00)
frequent	0.013	(0.02)	0.003	(0.00)	0.024	(0.02)
Urbanicity
never	0.003	(0.00)	0.014	(0.01)	0.010	(0.00) **
occasional	0.027	(0.01)	−0.007	(0.01)	0.027	(0.01) *
monthly	−0.010	(0.01)	−0.005	(0.00)	−0.005	(0.00) **
frequent	−0.021	(0.01)	−0.002	(0.00)	−0.032	(0.01) *

dy/dx are reported. Standard errors are in parenthesis. * p < 0.05, ** p < 0.01, *** p < 0.001. Note: Marginal effects are presented in decimals; for example, 0.01 corresponds to a 1 percentage point (pp) change in probability. The positive values in marginal effects indicate increases in the probability of the outcome category, while the negative values indicate decreases in the probability of the outcome category. Because predicted probabilities across all outcome categories sum to one, a change in an explanatory variable is associated with an increase in some outcome categories and a decrease in others. Given that 19% of schools report frequent bullying, 1 pp increase in the probability of frequent bullying represents about a 5% relative increase in the likelihood of frequent bullying. Given that 3% of schools report frequent sexual harassment, 1 pp increase in the probability of frequent sexual harassment represents about a 33% relative increase in the likelihood of frequent sexual harassment. Given that 26% of schools report frequent cyberbullying, 1 pp increase in the probability of frequent cyberbullying represents about a 4% relative increase in the likelihood of frequent bullying. Marginal effects were calculated after 59% of the SSOCS 2020 survey responses were retained. 41% of the SSOCS 2020 survey responses were excluded because they were completed after school closures in spring 2020.

for the results of this sensitivity analysis).

Guided by the socio-ecological framework, we hypothesize that model fit will improve as hierarchical blocks for each socio-ecological level are incorporated. At the mesosystem level, we expect that schools with more targeted programs (teacher training, school safety and climate efforts, and school discipline-related practices) will have lower rates of all three types of incidents. Also, structural features of the school environment such as being a middle or high school as opposed to an elementary school, and having a larger school size are expected to be associated with a higher frequency of incidents. At the exosystem level, higher neighborhood crime rates are expected to be associated with a higher frequency of incidents while greater community involvement (parents, juvenile justice and law enforcement, mental health and social services, and civic involvement) is expected to be associated with a lower frequency of incidents. At the macrosystem level, we expect schools located in rural areas as opposed to urban areas to have higher rates of bullying, sexual harassment, and cyberbullying, potentially due to more conservative political views in rural areas.

5. Results

5.1. Descriptive Statistics

Our sample included 5132 schools. Schools had varying levels of frequency for bullying, sexual harassment, and cyberbullying. 19% of schools had frequent bullying, 19% of schools had monthly bullying, 59% had occasional bullying, and 3% did not have any bullying incidents. 3% of schools had frequent sexual harassment, 6% of schools had monthly sexual harassment, 61% had occasional sexual harassment, and 30% did not have any sexual harassment. 26% of schools had frequent cyberbullying, 23% of schools had monthly cyberbullying, 45% had occasional cyberbullying, and 7% did not have any cyberbullying incidents. The average teacher training score is 9.75 out of 13, while the average school safety & climate score is 6.92 out of 11. Disciplinary practices varied across schools: 91% reported using security cameras, 65% had a threat reporting system, 58% had a threat assessment team, 68% had sworn law enforcement officers, and 9% performed regular or random metal detector checks. Parental and community involvement varied across schools: 28% of schools had high, 58% had moderate, and 14% had no parental involvement. 88% of schools involved juvenile justice and law enforcement agencies, 84% of schools involved mental health and social service agencies, and 62% of schools involved civic organizations, service clubs, private corporations, businesses, or religious organizations in schools’ efforts to promote safe schools. Schools in the sample consisted of grade level (25% elementary, 35% middle, 36% high/secondary, and 4% combined/other), enrollment size (11% enrolling fewer than 300 students, 22% enrolling 300–499 students, 39% enrolling 500–999 students, and 28% enrolling 1000 or more), and geographic spread (63% of schools were located in urban areas while 37% in rural areas). Descriptive statistics for all study variables are seen in

Table 1. Descriptive Statistics of Study Variables (N = 5132).

	Frequency	Percent
Dependent Variables
Bullying Frequency
(0 = never)	136	2.65%
(1 = occasional)	3045	59.33%
(2 = monthly)	983	19.15%
(3 = frequent)	968	18.86%
Sexual Harassment Frequency
(0 = never)	1518	29.58%
(1 = occasional)	3150	61.38%
(2 = monthly)	328	6.39%
(3 = frequent)	136	2.65%
Cyberbullying Frequency
(0 = never)	342	6.66%
(1 = occasional)	2285	44.52%
(2 = monthly)	1184	23.07%
(3 = frequent)	1321	25.74%
Mesosystem-Level Factors
Teacher Training Score	Mean 9.75; Std. dev. 3.22
School Safety & Climate Score	Mean 6.92; Std. dev. 2.06
School Disciplinary Practices
Security cameras (1 = Yes)	4681	91.21%
Threat reporting system (1 = Yes)	3317	64.63%
Threat assessment team (1 = Yes)	2971	57.89%
Sworn law enforcement officers (1 = Yes)	3473	67.67%
Metal detector checks (1 = Yes)	479	9.33%
Grade Level of School
(1 = elementary)	1295	25.23%
(2 = middle)	1793	34.94%
(3 = high/secondary school)	1824	35.54%
(4 = combined/other)	220	4.29%
School Size
(1 = school size less than 300)	548	10.68%
(2 = school size of 300–499)	1112	21.67%
(3 = school size of 500–999)	2016	39.28%
(4 = school size more than 1000)	1456	28.37%
Exosystem-Level Factors
Parental Involvement
(0 = none)	714	13.91%
(1 = moderate)	2976	57.99%
(2 = high)	1442	28.10%
Community Involvement
Juvenile justice & law enforcement (1 = Yes)	4510	87.88%
Mental health & social services (1 = Yes)	4333	84.43%
Civic & businesses & religious org. (1 = Yes)	3178	61.93%
Neighborhood Crime
(1 = low)	3793	73.91%
(2 = moderate)	1019	19.86%
(3 = high)	320	6.24%
Macrosystem-Level Factors
Urbanicity (1 = Yes)	3254	63.41%
City	1356	26.42%
Suburb	1898	36.98%
Urbanicity (0 = No)	1878	36.59%
Town	707	13.78%
Rural	1171	22.82%
Survey Wave
2018	2762	53.82%
2020	2370	46.18%

Note: All categorical variables and binary variables show the counts and percentages across categories. There are only two continuous variables (teacher training score and school safety & climate score), which are shown as means and standard deviations.

.

5.2. Zero Order Correlations

The Pearson correlations (

Table 2. Correlation Coefficients.

Pearson Correlations	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19
1. Bullying Frequency	1.0
2. Sexual Harassment Frequency	0.5	1.0
3. Cyberbullying Frequency	0.5	0.5	1.0
4. Teacher Training Score	−0.1	0.0	0.0	1.0
5. School Safety & Climate Score	0.0	0.1	0.1	0.3	1.0
6. Security cameras	0.0	0.0	0.1	0.1	0.0	1.0
7. Threat reporting system	0.0	0.1	0.1	0.2	0.2	0.1	1.0
8. Threat assessment team	0.0	0.1	0.1	0.3	0.3	0.0	0.2	1.0
9. Sworn law enforcement officers	0.1	0.1	0.2	0.1	0.1	0.2	0.2	0.1	1.0
10. Metal detector checks	0.0	0.0	0.0	0.1	0.1	0.1	0.1	0.0	0.1	1.0
11. Grade Level of School	0.1	0.2	0.3	0.0	0.1	0.2	0.1	0.0	0.2	0.2	1.0
12. School Size	0.1	0.2	0.2	0.1	0.3	0.1	0.2	0.1	0.3	0.1	0.3	1.0
13. Parental Involvement	0.0	0.0	−0.1	0.3	0.3	0.0	0.1	0.2	0.0	0.1	−0.1	0.1	1.0
14. Juvenile justice & law enforcement	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.1	0.2	0.0	0.2	0.1	0.1	1.0
15. Mental health & social services	0.1	0.1	0.1	0.2	0.2	0.1	0.1	0.2	0.1	0.0	0.1	0.1	0.2	0.3	1.0
16. Civic & businesses & religious org.	0.0	0.0	0.1	0.2	0.3	0.0	0.1	0.1	0.1	0.1	0.1	0.1	0.3	0.2	0.3	1.0
17. Neighborhood Crime	0.1	0.0	0.0	0.0	0.1	0.0	0.0	0.0	−0.1	0.2	0.0	0.0	0.1	−0.1	0.0	0.0	1.0
18. Urbanicity	0.0	0.0	0.0	0.1	0.3	−0.1	0.1	0.1	0.0	0.1	−0.1	0.3	0.1	−0.1	0.0	0.0	0.2	1.0
19. Survey Wave	0.0	0.0	0.0	0.1	0.1	0.1	0.2	0.2	0.0	0.0	0.0	0.0	0.1	0.1	0.1	0.0	0.0	0.0	1.0

) indicate that there is no evidence of multicollinearity among explanatory variables.

5.3. Generalized Ordered Logistics Models

In Table 3, Table 4 and Table 5, the first model analyzes the associations between mesosystem-level factors and the outcomes of interest. The second model adds exosystem-level factors in addition to the other variables used in Model 1. The third model adds macrosystem-level factors in addition to the other variables used in Model 2. The likelihood ratio (LR) test directly compares nested models to assess whether additional variables in the model significantly enhance the explanatory power of the model. Therefore, model fit is evaluated using the LR test, while Akaike Information Criterion (AIC), the Bayesian Information Criterion (BIC), and McFadden’s R-Squared are reported for transparency. A p-value of <0.05 in the LR test indicates that the additional variables provide a significant improvement in the model fit, while lower AIC and BIC values and higher McFadden’s R-Squared values indicate improved model fit.

For bullying, the inclusion of both exosystem-level variables (LR chi2(9) = 80.63, p-value = 0.0000) and macrosystem-level variables significantly improves model fit (LR chi2(4) = 10.11, p-value = 0.0386). For sexual harassment, the inclusion of exosystem-level variables significantly improves model fit (LR chi2(7) = 52.01, p-value = 0.0000) while the addition of macrosystem-level variables does not significantly improve model fit (LR chi2(2) = 1.05, p-value = 0.5926). For cyberbullying, the inclusion of both exosystem-level variables (LR chi2(7) = 89.66, p-value = 0.0000) and macrosystem-level variables significantly improves model fit (LR chi2(2) = 15.29, p-value = 0.0005). According to LR test results, the inclusion of exosystem-level variables significantly improves model fit for the three models, while the addition of macrosystem-level factors significantly increases the model fit for bullying and cyberbullying, but not for sexual harassment. Informed by the likelihood ratio test results and guided by the socio-ecological framework, we use Model 3 as a preferred model to make comparisons across these three behaviors.

Table 3, Table 4 and Table 5 show the estimates and their significance from generalized ordered logistic regression models. The summary table from these findings is also provided to compare the protective and risk factors across these three behaviors at each level (please refer to Table 6). Because coefficients from these tables are not directly interpretable, we report the direction and strength of associations between variables by using their signs and statistical significance. Positive coefficients indicate a higher likelihood of being in a higher-frequency category for the outcome variables (bullying, sexual harassment, and cyberbullying), while negative coefficients indicate a lower likelihood. Interpretation of effect sizes is provided through marginal effects (Table 7), which translates these coefficients into changes in predicted probabilities.

We find that more comprehensive teacher training and higher parental involvement are common protective factors for all three behaviors, significantly associated with reduced frequency of incidents (b coefficients for teacher training score and parental involvement are all negative in Table 3, Table 4 and Table 5). We also find that schools that have regular or random metal detector checks are more likely to have lower frequencies of bullying or sexual harassment compared to those schools that do not have such practices (Table 3, b coefficient for metal detector checks = −0.22, p-value = 0.031; Table 4, b coefficient for metal detector checks = −0.38, p-value = 0.001).

School grade level, enrollment size, neighborhood crime, and the community involvement of mental health & social services are common risk factors for all three behaviors. Being in higher grade levels compared to elementary school, larger school size, and higher neighborhood crime levels are significantly associated with increased frequency of incidents (b coefficients for school grade level, school size, and neighborhood crime are all positive in Table 3, Table 4 and Table 5). Schools that involve mental health agencies or social service agencies in their school safety efforts are more likely to have higher frequencies of incidents compared to schools without such partnerships (b coefficients for mental health & social services are all positive in Table 3, Table 4 and Table 5).

Sexual harassment and cyberbullying also have some distinct protective and risk factors, including the existence of security cameras for sexual harassment, and existence of threat reporting systems, sworn law enforcement, juvenile justice & law enforcement involvement, and urbanicity for cyberbullying. We find that schools with security cameras are more likely to have lower sexual harassment frequency compared to those without such practices (Table 4, b coefficient for security cameras = −0.52, p-value = 0.003). Schools that have threat reporting systems, sworn law enforcement officers, and community involvement for juvenile justice & law enforcement are more likely to have higher cyberbullying frequency compared to those without such practices (b coefficients for these variables are all positive in Table 5). Finally, schools that are located in urban areas (city or suburbs) are more likely than schools that are located in non-urban areas (town or rural) to have lower cyberbullying frequency (Table 5, b coefficient = −0.16, p-value = 0.014).

Table 3. Generalized Ordered Logistic Regression Results (DV = Bullying Frequency).

	Model 1		Model 2		Model 3
Mesosystem-Level Factors
Teacher Training Score	−0.06	(0.01) ***	−0.06	(0.01) ***	−0.06	(0.01) ***
School Safety & Climate Score	0.01	(0.02)	0.00	(0.02)	0.01	(0.02)
School Disciplinary Practices
Security cameras	−0.12	(0.10)	−0.14	(0.10)	−0.15	(0.10)
Threat reporting system	−0.03	(0.06)	−0.01	(0.06)	0.00	(0.06)
Threat assessment team	0.05	(0.06)	0.03	(0.06)	0.04	(0.06)
Sworn law enforcement officers	0.07	(0.07)	0.07	(0.07)	0.06	(0.07)
Metal detector checks	−0.14	(0.10)	−0.23	(0.11) *	−0.22	(0.11) *
Grade Level of School
(2 = middle)	1.04	(0.08) ***	1.02	(0.09) ***	1.01	(0.09) ***
(3 = high/secondary school)	0.43	(0.09) ***	0.41	(0.09) ***	0.38	(0.10) ***
(4 = combined/other)	0.34	(0.15) *	0.33	(0.15) *	0.30	(0.15)
School Size
(2 = school size of 300–499)	0.27	(0.12) *	0.31	(0.12) **	0.32	(0.12) **
(3 = school size of 500–999)	0.48	(0.11) ***	0.56	(0.11) ***	0.59	(0.11) ***
(4 = school size more than 1000)	0.80	(0.11) ***	0.86	(0.12) ***	0.91	(0.12) ***
Exosystem-Level Factors
Parental Involvement
(1 = moderate)			−0.17	(0.09)	−0.16	(0.09)
(2 = high)			−0.20	(0.11)	−0.19	(0.10) *
Community Involvement
Juvenile justice & law enforcement			0.10	(0.10)	0.09	(0.10)
Mental health & social services			0.46	(0.09) ***	0.46	(0.09) ***
Civic & businesses & religious org.			−0.04	(0.06)	−0.05	(0.06)
Neighborhood Crime
(2 = moderate)			0.30	(0.07) ***	0.32	(0.08) ***
(3 = high)			0.47	(0.13) ***	0.51	(0.13) ***
Macrosystem-Level Factors
Urbanicity					−0.11	(0.07)
Survey Wave (2020)					−0.10	(0.06)
McFadden’s R-Squared	0.0378		0.0453		0.0463
AIC	10,295		10,232		10,230
BIC	10,465		10,461		10,485

Dependent variable is bullying frequency. Unstandardized coefficients are reported. Standard errors are in parenthesis. * p < 0.05, ** p < 0.01, *** p < 0.001. Note: Estimates are from generalized ordered logistic regression models. The Wald test result indicates no significant violation of the proportional odds assumption after partial constraints were applied (chi2(30) = 27.32; Prob > chi2 = 0.6064). For simplicity, the coefficients reported in this table correspond to the medium threshold (transition from the occasional category to the monthly category). For bullying, the following predictors showed varying coefficients across thresholds: teacher training, grade level, school size, parental involvement, and neighborhood crime level. Full results across thresholds are reflected in the marginal effects (Table 7), which provide interpretable estimates of changes in predicted probabilities.

Table 3. Generalized Ordered Logistic Regression Results (DV = Bullying Frequency).

	Model 1		Model 2		Model 3
Mesosystem-Level Factors
Teacher Training Score	−0.06	(0.01) ***	−0.06	(0.01) ***	−0.06	(0.01) ***
School Safety & Climate Score	0.01	(0.02)	0.00	(0.02)	0.01	(0.02)
School Disciplinary Practices
Security cameras	−0.12	(0.10)	−0.14	(0.10)	−0.15	(0.10)
Threat reporting system	−0.03	(0.06)	−0.01	(0.06)	0.00	(0.06)
Threat assessment team	0.05	(0.06)	0.03	(0.06)	0.04	(0.06)
Sworn law enforcement officers	0.07	(0.07)	0.07	(0.07)	0.06	(0.07)
Metal detector checks	−0.14	(0.10)	−0.23	(0.11) *	−0.22	(0.11) *
Grade Level of School
(2 = middle)	1.04	(0.08) ***	1.02	(0.09) ***	1.01	(0.09) ***
(3 = high/secondary school)	0.43	(0.09) ***	0.41	(0.09) ***	0.38	(0.10) ***
(4 = combined/other)	0.34	(0.15) *	0.33	(0.15) *	0.30	(0.15)
School Size
(2 = school size of 300–499)	0.27	(0.12) *	0.31	(0.12) **	0.32	(0.12) **
(3 = school size of 500–999)	0.48	(0.11) ***	0.56	(0.11) ***	0.59	(0.11) ***
(4 = school size more than 1000)	0.80	(0.11) ***	0.86	(0.12) ***	0.91	(0.12) ***
Exosystem-Level Factors
Parental Involvement
(1 = moderate)			−0.17	(0.09)	−0.16	(0.09)
(2 = high)			−0.20	(0.11)	−0.19	(0.10) *
Community Involvement
Juvenile justice & law enforcement			0.10	(0.10)	0.09	(0.10)
Mental health & social services			0.46	(0.09) ***	0.46	(0.09) ***
Civic & businesses & religious org.			−0.04	(0.06)	−0.05	(0.06)
Neighborhood Crime
(2 = moderate)			0.30	(0.07) ***	0.32	(0.08) ***
(3 = high)			0.47	(0.13) ***	0.51	(0.13) ***
Macrosystem-Level Factors
Urbanicity					−0.11	(0.07)
Survey Wave (2020)					−0.10	(0.06)
McFadden’s R-Squared	0.0378		0.0453		0.0463
AIC	10,295		10,232		10,230
BIC	10,465		10,461		10,485

Dependent variable is bullying frequency. Unstandardized coefficients are reported. Standard errors are in parenthesis. * p < 0.05, ** p < 0.01, *** p < 0.001. Note: Estimates are from generalized ordered logistic regression models. The Wald test result indicates no significant violation of the proportional odds assumption after partial constraints were applied (chi2(30) = 27.32; Prob > chi2 = 0.6064). For simplicity, the coefficients reported in this table correspond to the medium threshold (transition from the occasional category to the monthly category). For bullying, the following predictors showed varying coefficients across thresholds: teacher training, grade level, school size, parental involvement, and neighborhood crime level. Full results across thresholds are reflected in the marginal effects (Table 7), which provide interpretable estimates of changes in predicted probabilities.

Table 4. Generalized Ordered Logistic Regression Results (DV = Sexual Harassment Frequency).

	Model 1		Model 2		Model 3
Mesosystem-Level Factors
Teacher Training Score	−0.03	(0.01) **	−0.03	(0.01) *	−0.03	(0.01) *
School Safety & Climate Score	0.04	(0.02) *	0.03	(0.02)	0.03	(0.02)
School Disciplinary Practices
Security cameras	−0.12	(0.11)	−0.15	(0.11)	−0.52	(0.17) **
Threat reporting system	0.06	(0.07)	0.08	(0.07)	0.08	(0.07)
Threat assessment team	−0.10	(0.10)	0.09	(0.07)	−0.12	(0.10)
Sworn law enforcement officers	−0.01	(0.07)	−0.02	(0.07)	−0.02	(0.07)
Metal detector checks	−0.31	(0.11) **	−0.38	(0.11) **	−0.38	(0.11) **
Grade Level of School
(2 = middle)	1.69	(0.09) ***	1.66	(0.09) ***	2.08	(0.13) ***
(3 = high/secondary school)	1.26	(0.13) ***	1.20	(0.13) ***	1.66	(0.10) ***
(4 = combined/other)	0.88	(0.15) ***	0.88	(0.15) ***	0.90	(0.16) ***
School Size
(2 = school size of 300–499)	0.47	(0.12) ***	0.45	(0.12) ***	0.46	(0.12) ***
(3 = school size of 500–999)	0.68	(0.11) ***	0.69	(0.11) ***	0.70	(0.11) ***
(4 = school size more than 1000)	1.18	(0.12) ***	1.22	(0.12) ***	1.23	(0.13) ***
Exosystem-Level Factors
Parental Involvement
(1 = moderate)			−0.11	(0.09)	−0.10	(0.09)
(2 = high)			−0.35	(0.14) *	−0.33	(0.14) *
Community Involvement
Juvenile justice & law enforcement			0.13	(0.10)	0.12	(0.10)
Mental health & social services			0.35	(0.09) ***	0.34	(0.09) ***
Civic & businesses & religious org.			0.01	(0.07)	0.01	(0.07)
Neighborhood Crime
(2 = moderate)			0.27	(0.08) ***	0.28	(0.08) ***
(3 = high)			0.50	(0.14) ***	0.51	(0.14) ***
Macrosystem-Level Factors
Urbanicity					−0.06	(0.07)
Survey Wave (2020)					0.04	(0.06)
McFadden’s R-Squared	0.1007		0.1062		0.1073
AIC	8641		8603		8605
BIC	8772		8780		8821

Dependent variable is sexual harassment frequency. Unstandardized coefficients are reported. Standard errors are in parenthesis. * p < 0.05, ** p < 0.01, *** p < 0.001. Note: Estimates are from generalized ordered logistic regression models. The Wald test result indicates no significant violation of the proportional odds assumption after partial constraints were applied (chi2(36) = 51.68, Prob > chi2 = 0.0498). For simplicity, the coefficients reported in this table correspond to the medium threshold (transition from the occasional category to the monthly category). For sexual harassment, the following predictors showed varying coefficients across thresholds: security cameras, threat assessment team, grade level, and parental involvement. Full results across thresholds are reflected in the marginal effects (Table 7), which provide interpretable estimates of changes in predicted probabilities.

Table 4. Generalized Ordered Logistic Regression Results (DV = Sexual Harassment Frequency).

	Model 1		Model 2		Model 3
Mesosystem-Level Factors
Teacher Training Score	−0.03	(0.01) **	−0.03	(0.01) *	−0.03	(0.01) *
School Safety & Climate Score	0.04	(0.02) *	0.03	(0.02)	0.03	(0.02)
School Disciplinary Practices
Security cameras	−0.12	(0.11)	−0.15	(0.11)	−0.52	(0.17) **
Threat reporting system	0.06	(0.07)	0.08	(0.07)	0.08	(0.07)
Threat assessment team	−0.10	(0.10)	0.09	(0.07)	−0.12	(0.10)
Sworn law enforcement officers	−0.01	(0.07)	−0.02	(0.07)	−0.02	(0.07)
Metal detector checks	−0.31	(0.11) **	−0.38	(0.11) **	−0.38	(0.11) **
Grade Level of School
(2 = middle)	1.69	(0.09) ***	1.66	(0.09) ***	2.08	(0.13) ***
(3 = high/secondary school)	1.26	(0.13) ***	1.20	(0.13) ***	1.66	(0.10) ***
(4 = combined/other)	0.88	(0.15) ***	0.88	(0.15) ***	0.90	(0.16) ***
School Size
(2 = school size of 300–499)	0.47	(0.12) ***	0.45	(0.12) ***	0.46	(0.12) ***
(3 = school size of 500–999)	0.68	(0.11) ***	0.69	(0.11) ***	0.70	(0.11) ***
(4 = school size more than 1000)	1.18	(0.12) ***	1.22	(0.12) ***	1.23	(0.13) ***
Exosystem-Level Factors
Parental Involvement
(1 = moderate)			−0.11	(0.09)	−0.10	(0.09)
(2 = high)			−0.35	(0.14) *	−0.33	(0.14) *
Community Involvement
Juvenile justice & law enforcement			0.13	(0.10)	0.12	(0.10)
Mental health & social services			0.35	(0.09) ***	0.34	(0.09) ***
Civic & businesses & religious org.			0.01	(0.07)	0.01	(0.07)
Neighborhood Crime
(2 = moderate)			0.27	(0.08) ***	0.28	(0.08) ***
(3 = high)			0.50	(0.14) ***	0.51	(0.14) ***
Macrosystem-Level Factors
Urbanicity					−0.06	(0.07)
Survey Wave (2020)					0.04	(0.06)
McFadden’s R-Squared	0.1007		0.1062		0.1073
AIC	8641		8603		8605
BIC	8772		8780		8821

Dependent variable is sexual harassment frequency. Unstandardized coefficients are reported. Standard errors are in parenthesis. * p < 0.05, ** p < 0.01, *** p < 0.001. Note: Estimates are from generalized ordered logistic regression models. The Wald test result indicates no significant violation of the proportional odds assumption after partial constraints were applied (chi2(36) = 51.68, Prob > chi2 = 0.0498). For simplicity, the coefficients reported in this table correspond to the medium threshold (transition from the occasional category to the monthly category). For sexual harassment, the following predictors showed varying coefficients across thresholds: security cameras, threat assessment team, grade level, and parental involvement. Full results across thresholds are reflected in the marginal effects (Table 7), which provide interpretable estimates of changes in predicted probabilities.

Table 5. Generalized Ordered Logistic Regression Results (DV = Cyberbullying Frequency).

	Model 1		Model 2		Model 3
Mesosystem-Level Factors
Teacher Training Score	−0.03	(0.01) **	−0.03	(0.01) **	−0.03	(0.01) **
School Safety & Climate Score	−0.02	(0.02)	−0.02	(0.02)	−0.01	(0.02)
School Disciplinary Practices
Security cameras	0.12	(0.10)	0.08	(0.10)	0.03	(0.10)
Threat reporting system	0.17	(0.06) **	0.19	(0.06) **	0.18	(0.06) **
Threat assessment team	0.06	(0.06)	0.04	(0.06)	0.03	(0.06)
Sworn law enforcement officers	0.22	(0.06) ***	0.20	(0.06) **	0.19	(0.06) **
Metal detector checks	−0.11	(0.09)	−0.16	(0.10)	−0.15	(0.10)
Grade Level of School
(2 = middle)	2.05	(0.09) ***	2.01	(0.09) ***	2.00	(0.09) ***
(3 = high/secondary school)	1.77	(0.10) ***	1.68	(0.10) ***	1.65	(0.10) ***
(4 = combined/other)	1.02	(0.15) ***	1.00	(0.15) ***	0.94	(0.15) ***
School Size
(2 = school size of 300–499)	0.58	(0.11) ***	0.56	(0.11) ***	0.60	(0.11) ***
(3 = school size of 500–999)	0.73	(0.10) ***	0.74	(0.10) ***	0.80	(0.11) ***
(4 = school size more than 1000)	0.98	(0.11) ***	1.03	(0.11) ***	1.12	(0.12) ***
Exosystem-Level Factors
Parental Involvement
(1 = moderate)			−0.17	(0.08) *	−0.16	(0.08)
(2 = high)			−0.44	(0.10) ***	−0.43	(0.10) ***
Community Involvement
Juvenile justice & law enforcement			0.20	(0.09) *	0.19	(0.09) *
Mental health & social services			0.43	(0.08) ***	0.41	(0.08) ***
Civic & businesses & religious org.			0.08	(0.06)	0.07	(0.06)
Neighborhood Crime
(2 = moderate)			0.31	(0.07) ***	0.34	(0.07) ***
(3 = high)			0.40	(0.12) **	0.45	(0.12) ***
Macrosystem-Level Factors
Urbanicity					−0.16	(0.06) *
Survey Wave (2020)					0.05	(0.06)
McFadden’s R-Squared	0.1015		0.1086		0.1098
AIC	11,373		11,298		11,286
BIC	11,517		11,487		11,489

Dependent variable is cyberbullying frequency. Unstandardized coefficients are reported. Standard errors are in parenthesis. * p < 0.05, ** p < 0.01, *** p < 0.001. Note: Estimates are from generalized ordered logistic regression models. The Wald test result indicates no significant violation of the proportional odds assumption after partial constraints were applied (chi2(38) = 44.38, Prob > chi2 = 0.2206). For simplicity, the coefficients reported in this table correspond to the medium threshold (transition from the occasional category to the monthly category). For cyberbullying, the grade level showed varying coefficients across thresholds. Full results across thresholds are reflected in the marginal effects (Table 7), which provide interpretable estimates of changes in predicted probabilities.

Table 5. Generalized Ordered Logistic Regression Results (DV = Cyberbullying Frequency).

	Model 1		Model 2		Model 3
Mesosystem-Level Factors
Teacher Training Score	−0.03	(0.01) **	−0.03	(0.01) **	−0.03	(0.01) **
School Safety & Climate Score	−0.02	(0.02)	−0.02	(0.02)	−0.01	(0.02)
School Disciplinary Practices
Security cameras	0.12	(0.10)	0.08	(0.10)	0.03	(0.10)
Threat reporting system	0.17	(0.06) **	0.19	(0.06) **	0.18	(0.06) **
Threat assessment team	0.06	(0.06)	0.04	(0.06)	0.03	(0.06)
Sworn law enforcement officers	0.22	(0.06) ***	0.20	(0.06) **	0.19	(0.06) **
Metal detector checks	−0.11	(0.09)	−0.16	(0.10)	−0.15	(0.10)
Grade Level of School
(2 = middle)	2.05	(0.09) ***	2.01	(0.09) ***	2.00	(0.09) ***
(3 = high/secondary school)	1.77	(0.10) ***	1.68	(0.10) ***	1.65	(0.10) ***
(4 = combined/other)	1.02	(0.15) ***	1.00	(0.15) ***	0.94	(0.15) ***
School Size
(2 = school size of 300–499)	0.58	(0.11) ***	0.56	(0.11) ***	0.60	(0.11) ***
(3 = school size of 500–999)	0.73	(0.10) ***	0.74	(0.10) ***	0.80	(0.11) ***
(4 = school size more than 1000)	0.98	(0.11) ***	1.03	(0.11) ***	1.12	(0.12) ***
Exosystem-Level Factors
Parental Involvement
(1 = moderate)			−0.17	(0.08) *	−0.16	(0.08)
(2 = high)			−0.44	(0.10) ***	−0.43	(0.10) ***
Community Involvement
Juvenile justice & law enforcement			0.20	(0.09) *	0.19	(0.09) *
Mental health & social services			0.43	(0.08) ***	0.41	(0.08) ***
Civic & businesses & religious org.			0.08	(0.06)	0.07	(0.06)
Neighborhood Crime
(2 = moderate)			0.31	(0.07) ***	0.34	(0.07) ***
(3 = high)			0.40	(0.12) **	0.45	(0.12) ***
Macrosystem-Level Factors
Urbanicity					−0.16	(0.06) *
Survey Wave (2020)					0.05	(0.06)
McFadden’s R-Squared	0.1015		0.1086		0.1098
AIC	11,373		11,298		11,286
BIC	11,517		11,487		11,489

Dependent variable is cyberbullying frequency. Unstandardized coefficients are reported. Standard errors are in parenthesis. * p < 0.05, ** p < 0.01, *** p < 0.001. Note: Estimates are from generalized ordered logistic regression models. The Wald test result indicates no significant violation of the proportional odds assumption after partial constraints were applied (chi2(38) = 44.38, Prob > chi2 = 0.2206). For simplicity, the coefficients reported in this table correspond to the medium threshold (transition from the occasional category to the monthly category). For cyberbullying, the grade level showed varying coefficients across thresholds. Full results across thresholds are reflected in the marginal effects (Table 7), which provide interpretable estimates of changes in predicted probabilities.

Table 6. Summary Table.

	Protective	Risk	Form of Behavior
Mesosystem-Level Factors
Teacher Training Score	X		B, SH, CB
School Disciplinary Practices
Security cameras	X		SH
Threat reporting system		X	CB
Sworn law enforcement officers		X	CB
Metal detector checks	X		B, SH
Grade Level of School		X	B, SH, CB
School Enrollment Size		X	B, SH, CB
Exosystem-Level Factors
Parental Involvement	X		B, SH, CB
Community Involvement of Juvenile justice & law enforcement		X	CB
Community Involvement of Mental health & social services		X	B, SH, CB
Neighborhood Crime		X	B, SH, CB
Macrosystem-Level Factors
Urbanicity	X		CB

Note: The table provides a summary of Table 3, Table 4 and Table 5. Significant protective and risk factors for different forms of behavior are shown. B refers to bullying, SH refers to sexual harassment, and CB refers to cyberbullying.

Table 6. Summary Table.

	Protective	Risk	Form of Behavior
Mesosystem-Level Factors
Teacher Training Score	X		B, SH, CB
School Disciplinary Practices
Security cameras	X		SH
Threat reporting system		X	CB
Sworn law enforcement officers		X	CB
Metal detector checks	X		B, SH
Grade Level of School		X	B, SH, CB
School Enrollment Size		X	B, SH, CB
Exosystem-Level Factors
Parental Involvement	X		B, SH, CB
Community Involvement of Juvenile justice & law enforcement		X	CB
Community Involvement of Mental health & social services		X	B, SH, CB
Neighborhood Crime		X	B, SH, CB
Macrosystem-Level Factors
Urbanicity	X		CB

Note: The table provides a summary of Table 3, Table 4 and Table 5. Significant protective and risk factors for different forms of behavior are shown. B refers to bullying, SH refers to sexual harassment, and CB refers to cyberbullying.

Table 7. Marginal Effects of Significant Study Variables.

	Bullying Frequency		Sexual Harassment Frequency		Cyberbullying Frequency
Common Protective Factors
Teacher Training Score
never	−0.001	(0.00)	0.004	(0.00) *	0.002	(0.00) **
occasional	0.014	(0.00) ***	−0.002	(0.00) *	0.005	(0.00) **
monthly	−0.005	(0.00) **	−0.001	(0.00) *	−0.001	(0.00) **
frequent	−0.008	(0.00) ***	−0.001	(0.00) *	−0.006	(0.00) **
High Parental Involvement (compared to no parental involvement)
never	0.013	(0.01) *	0.054	(0.02) **	0.023	(0.01) ***
occasional	0.030	(0.02)	−0.029	(0.02)	0.065	(0.02) ***
monthly	0.012	(0.01)	−0.028	(0.01) **	−0.014	(0.00) ***
frequent	−0.054	(0.02) **	0.003	0.006	−0.074	0.017***
Existence of Metal Detector Checks
never	0.006	(0.00)	0.067	(0.02) **	0.008	(0.01)
occasional	0.041	(0.02) *	−0.040	(0.01) **	0.022	(0.01)
monthly	−0.017	(0.01) *	−0.018	(0.00) ***	−0.005	(0.00)
frequent	−0.030	(0.01) *	−0.008	(0.00) ***	−0.025	(0.02)
Common Risk Factors
Middle School (compared to Elementary School)
never	−0.039	(0.01) ***	−0.342	(0.02) ***	−0.159	(0.01) ***
occasional	−0.185	(0.02) ***	0.225	(0.02) ***	−0.271	(0.02) ***
monthly	0.056	(0.01) ***	0.080	(0.01) ***	0.162	(0.01) ***
frequent	0.168	(0.01) ***	0.036	(0.00) ***	0.267	(0.01) ***
High/Secondary School (compared to Elementary School)
never	−0.035	(0.01) ***	−0.346	(0.02) ***	−0.151	(0.01) ***
occasional	−0.044	(0.02) *	0.272	(0.02) ***	−0.198	(0.02) ***
monthly	0.034	(0.01) *	0.053	(0.00) ***	0.137	(0.01) ***
frequent	0.046	(0.01) **	0.021	(0.00) ***	0.213	(0.01) ***
School Size more than 1000 (compared to school size less than 300)
never	−0.028	(0.01) ***	−0.225	(0.02) ***	−0.066	(0.01) ***
occasional	−0.166	(0.02) ***	0.138	(0.02) ***	−0.167	(0.02) ***
monthly	0.071	(0.01) ***	0.059	(0.01) ***	0.057	(0.01) ***
frequent	0.122	(0.01) ***	0.027	(0.00) ***	0.175	(0.02) ***
High Neighborhood Crime (compared to low neighborhood crime)
never	0.000	(0.01)	−0.083	(0.02) ***	−0.022	(0.01) ***
occasional	−0.115	(0.03) ***	0.037	(0.01) ***	−0.068	(0.02) ***
monthly	−0.025	(0.02)	0.030	(0.01) **	0.010	(0.00) ***
frequent	0.140	(0.03) ***	0.015	(0.01) **	0.080	(0.02) **
Existence of Mental Health & Social Services Involvement
never	−0.013	(0.00) ***	−0.060	(0.02) ***	−0.024	0.005***
occasional	−0.084	(0.01) ***	0.036	(0.01) **	−0.061	0.012***
monthly	0.037	(0.01) ***	0.016	(0.00) ***	0.018	(0.00) ***
frequent	0.061	(0.01) ***	0.008	(0.00) ***	0.066	(0.01) ***
Distinct Protective & Risk Factors
Existence of Security Cameras
never	0.004	(0.00)	0.009	(0.02)	−0.002	(0.01)
occasional	0.030	(0.02)	0.039	(0.02)	−0.005	(0.02)
monthly	−0.011	(0.01)	−0.041	(0.02) *	0.001	(0.00)
frequent	−0.023	(0.02)	−0.007	(0.01)	0.006	(0.02)
Existence of Threat Reporting System
never	0.000	(0.00)	−0.013	(0.01)	−0.010	(0.00) **
occasional	−0.001	(0.01)	0.007	(0.01)	−0.028	(0.01) **
monthly	0.000	(0.00)	0.004	(0.00)	0.006	(0.00) **
frequent	0.000	(0.01)	0.002	(0.00)	0.031	(0.01) **
Existence of Sworn Law Enforcement Officers
never	−0.002	(0.00)	0.004	(0.01)	−0.010	(0.00) **
occasional	−0.012	(0.01)	−0.002	(0.01)	−0.029	(0.01) **
monthly	0.004	(0.01)	−0.001	(0.00)	0.007	(0.00) *
frequent	0.009	(0.01)	−0.001	(0.00)	0.032	(0.01) **
Existence of Juvenile Justice & Law Enforcement Involvement
never	−0.002	(0.00)	−0.021	(0.02)	−0.010	(0.01)
occasional	−0.018	(0.02)	0.012	(0.01)	−0.028	(0.01) *
monthly	0.007	(0.01)	0.006	(0.00)	0.007	(0.00)
frequent	0.013	(0.01)	0.003	(0.00)	0.031	(0.01) *
Urbanicity
never	0.003	(0.00)	0.010	(0.01)	0.008	(0.00) *
occasional	0.021	(0.01)	−0.005	(0.01)	0.024	(0.01) *
monthly	−0.008	(0.00)	−0.003	(0.00)	−0.005	(0.00) *
frequent	−0.015	(0.01)	−0.001	(0.00)	−0.027	(0.01) *

dy/dx are reported. Standard errors are in parenthesis. * p < 0.05, ** p < 0.01, *** p < 0.001. Note: Marginal effects are presented in decimals; for example, 0.01 corresponds to a 1 percentage point (pp) change in probability. The positive values in marginal effects indicate increases in the probability of the outcome category, while the negative values indicate decreases in the probability of the outcome category. Because predicted probabilities across all outcome categories sum to one, a change in an explanatory variable is associated with an increase in some outcome categories and a decrease in others. Given that 19% of schools report frequent bullying, 1 pp increase in the probability of frequent bullying represents about a 5% relative increase in the likelihood of frequent bullying. Given that 3% of schools report frequent sexual harassment, 1 pp increase in the probability of frequent sexual harassment represents about a 33% relative increase in the likelihood of frequent sexual harassment. Given that 26% of schools report frequent cyberbullying, 1 pp increase in the probability of frequent cyberbullying represents about a 4% relative increase in the likelihood of frequent bullying.

Table 7. Marginal Effects of Significant Study Variables.

	Bullying Frequency		Sexual Harassment Frequency		Cyberbullying Frequency
Common Protective Factors
Teacher Training Score
never	−0.001	(0.00)	0.004	(0.00) *	0.002	(0.00) **
occasional	0.014	(0.00) ***	−0.002	(0.00) *	0.005	(0.00) **
monthly	−0.005	(0.00) **	−0.001	(0.00) *	−0.001	(0.00) **
frequent	−0.008	(0.00) ***	−0.001	(0.00) *	−0.006	(0.00) **
High Parental Involvement (compared to no parental involvement)
never	0.013	(0.01) *	0.054	(0.02) **	0.023	(0.01) ***
occasional	0.030	(0.02)	−0.029	(0.02)	0.065	(0.02) ***
monthly	0.012	(0.01)	−0.028	(0.01) **	−0.014	(0.00) ***
frequent	−0.054	(0.02) **	0.003	0.006	−0.074	0.017***
Existence of Metal Detector Checks
never	0.006	(0.00)	0.067	(0.02) **	0.008	(0.01)
occasional	0.041	(0.02) *	−0.040	(0.01) **	0.022	(0.01)
monthly	−0.017	(0.01) *	−0.018	(0.00) ***	−0.005	(0.00)
frequent	−0.030	(0.01) *	−0.008	(0.00) ***	−0.025	(0.02)
Common Risk Factors
Middle School (compared to Elementary School)
never	−0.039	(0.01) ***	−0.342	(0.02) ***	−0.159	(0.01) ***
occasional	−0.185	(0.02) ***	0.225	(0.02) ***	−0.271	(0.02) ***
monthly	0.056	(0.01) ***	0.080	(0.01) ***	0.162	(0.01) ***
frequent	0.168	(0.01) ***	0.036	(0.00) ***	0.267	(0.01) ***
High/Secondary School (compared to Elementary School)
never	−0.035	(0.01) ***	−0.346	(0.02) ***	−0.151	(0.01) ***
occasional	−0.044	(0.02) *	0.272	(0.02) ***	−0.198	(0.02) ***
monthly	0.034	(0.01) *	0.053	(0.00) ***	0.137	(0.01) ***
frequent	0.046	(0.01) **	0.021	(0.00) ***	0.213	(0.01) ***
School Size more than 1000 (compared to school size less than 300)
never	−0.028	(0.01) ***	−0.225	(0.02) ***	−0.066	(0.01) ***
occasional	−0.166	(0.02) ***	0.138	(0.02) ***	−0.167	(0.02) ***
monthly	0.071	(0.01) ***	0.059	(0.01) ***	0.057	(0.01) ***
frequent	0.122	(0.01) ***	0.027	(0.00) ***	0.175	(0.02) ***
High Neighborhood Crime (compared to low neighborhood crime)
never	0.000	(0.01)	−0.083	(0.02) ***	−0.022	(0.01) ***
occasional	−0.115	(0.03) ***	0.037	(0.01) ***	−0.068	(0.02) ***
monthly	−0.025	(0.02)	0.030	(0.01) **	0.010	(0.00) ***
frequent	0.140	(0.03) ***	0.015	(0.01) **	0.080	(0.02) **
Existence of Mental Health & Social Services Involvement
never	−0.013	(0.00) ***	−0.060	(0.02) ***	−0.024	0.005***
occasional	−0.084	(0.01) ***	0.036	(0.01) **	−0.061	0.012***
monthly	0.037	(0.01) ***	0.016	(0.00) ***	0.018	(0.00) ***
frequent	0.061	(0.01) ***	0.008	(0.00) ***	0.066	(0.01) ***
Distinct Protective & Risk Factors
Existence of Security Cameras
never	0.004	(0.00)	0.009	(0.02)	−0.002	(0.01)
occasional	0.030	(0.02)	0.039	(0.02)	−0.005	(0.02)
monthly	−0.011	(0.01)	−0.041	(0.02) *	0.001	(0.00)
frequent	−0.023	(0.02)	−0.007	(0.01)	0.006	(0.02)
Existence of Threat Reporting System
never	0.000	(0.00)	−0.013	(0.01)	−0.010	(0.00) **
occasional	−0.001	(0.01)	0.007	(0.01)	−0.028	(0.01) **
monthly	0.000	(0.00)	0.004	(0.00)	0.006	(0.00) **
frequent	0.000	(0.01)	0.002	(0.00)	0.031	(0.01) **
Existence of Sworn Law Enforcement Officers
never	−0.002	(0.00)	0.004	(0.01)	−0.010	(0.00) **
occasional	−0.012	(0.01)	−0.002	(0.01)	−0.029	(0.01) **
monthly	0.004	(0.01)	−0.001	(0.00)	0.007	(0.00) *
frequent	0.009	(0.01)	−0.001	(0.00)	0.032	(0.01) **
Existence of Juvenile Justice & Law Enforcement Involvement
never	−0.002	(0.00)	−0.021	(0.02)	−0.010	(0.01)
occasional	−0.018	(0.02)	0.012	(0.01)	−0.028	(0.01) *
monthly	0.007	(0.01)	0.006	(0.00)	0.007	(0.00)
frequent	0.013	(0.01)	0.003	(0.00)	0.031	(0.01) *
Urbanicity
never	0.003	(0.00)	0.010	(0.01)	0.008	(0.00) *
occasional	0.021	(0.01)	−0.005	(0.01)	0.024	(0.01) *
monthly	−0.008	(0.00)	−0.003	(0.00)	−0.005	(0.00) *
frequent	−0.015	(0.01)	−0.001	(0.00)	−0.027	(0.01) *

dy/dx are reported. Standard errors are in parenthesis. * p < 0.05, ** p < 0.01, *** p < 0.001. Note: Marginal effects are presented in decimals; for example, 0.01 corresponds to a 1 percentage point (pp) change in probability. The positive values in marginal effects indicate increases in the probability of the outcome category, while the negative values indicate decreases in the probability of the outcome category. Because predicted probabilities across all outcome categories sum to one, a change in an explanatory variable is associated with an increase in some outcome categories and a decrease in others. Given that 19% of schools report frequent bullying, 1 pp increase in the probability of frequent bullying represents about a 5% relative increase in the likelihood of frequent bullying. Given that 3% of schools report frequent sexual harassment, 1 pp increase in the probability of frequent sexual harassment represents about a 33% relative increase in the likelihood of frequent sexual harassment. Given that 26% of schools report frequent cyberbullying, 1 pp increase in the probability of frequent cyberbullying represents about a 4% relative increase in the likelihood of frequent bullying.

5.4. Marginal Effects

Table 7 presents the marginal effects of significant study variables. Below, we highlight the interpretation of select findings. More comprehensive teacher training and higher parental involvement are common protective factors for all three behaviors. An increase in teacher training score is associated with a lower likelihood of frequent incidents and a slightly higher likelihood of occasional or no incidents across all three behaviors. Specifically, a one-point increase in teacher training score is, on average, associated with

• 0.8 percentage point (pp) decrease in the probability of frequent bullying and 0.5 pp decrease in the probability of monthly bullying, with a corresponding increase in the probability of occasional bullying (1.4 pp).
• a decrease in the probabilities of frequent sexual harassment (0.1 pp), monthly sexual harassment (0.1 pp), and occasional sexual harassment (0.2 pp), with a corresponding increase in the probability of no sexual harassment (0.4).
• 0.6 pp decrease in the probability of frequent cyberbullying and 0.1 pp decrease in the probability of monthly cyberbullying, with a slight increase in occasional cyberbullying (0.5 pp) and no cyberbullying (0.2 pp).

Compared to schools with no parental involvement, those with high parental involvement are estimated to have

• 5.4 pp lower probability of frequent bullying, with a 1.3 pp higher probability of no bullying.
• 2.8 pp lower probability of monthly sexual harassment, with a 5.4 pp higher probability of no sexual harassment.
• 7.4 pp lower probability of frequent cyberbullying and 1.4 pp lower probability of monthly cyberbullying, with a higher probability of occasional cyberbullying (6.5 pp), and no cyberbullying (2.3 pp).

Existence of metal detector checks is a common protective factor for bullying and sexual harassment, while it is not a significant predictor for cyberbullying. Schools with random or regular metal detector checks compared to those without such practices are estimated to have

• 3 pp decrease in the probability of frequent bullying and 1.7 pp decrease in the probability of monthly bullying, with a corresponding increase in the probability of occasional bullying (4.1 pp) and no bullying (0.6 pp).
• a decrease in the probabilities of frequent sexual harassment (0.8 pp), monthly sexual harassment (1.8 pp), and occasional sexual harassment (4 pp), with a corresponding increase in the probability of no sexual harassment (6.7).

School grade level, enrollment size, neighborhood crime, and the community involvement of mental health & social services are common risk factors for all three behaviors. Among all predictors, school grade level has the strongest association with the frequency of bullying, sexual harassment, and cyberbullying. Middle schools and high/secondary schools are more likely than elementary schools to have frequent and monthly incidents and less likely to have no incidents across all three behaviors. To illustrate, compared to elementary schools, middle schools are estimated to have

• 16.8 pp higher probability of frequent bullying and 5.6 pp higher probability of monthly bullying, with a lower probability of occasional bullying (18.5 pp), and no bullying (3.9 pp).
• a higher probability of frequent sexual harassment (3.6 pp), monthly sexual harassment (8 pp), and occasional sexual harassment (22.5 pp), along with a 34.2 pp lower probability of no sexual harassment.
• 26.7 pp higher probability of frequent cyberbullying and 16.2 pp higher probability of monthly cyberbullying, with a lower probability of occasional cyberbullying (27.1 pp), and no cyberbullying (15.9 pp).

Sexual harassment and cyberbullying also have some distinct protective and risk factors, such that the existence of security cameras is a protective factor for sexual harassment, the existence of threat reporting systems, sworn law enforcement, and juvenile justice & law enforcement involvement are risk factors for cyberbullying, and urbanicity is a protective factor for cyberbullying. For example, schools in urban areas (city or suburban) are estimated to have lower probability of frequent cyberbullying (2.7 pp) and monthly cyberbullying (0.5 pp) and a higher probability of occasional cyberbullying (2.5 pp), and no cyberbullying (1 pp) compared to schools in non-urban (town or rural) areas.

6. Discussion and Conclusion

In this section, we present and discuss the findings as aligned with the levels of the socio-ecological model and compare and contrast the risk and protective factors for the three forms of peer harassment. We distinguish the findings that support the existing literature from those that challenge or add to the literature. As stated above, this dataset did not allow for examination of the microsystem factors that much of the literature has found to influence bullying. Students’ interactions with their peers, teachers, and parents (for example, see D’Urso et al., 2021) are important influences on their behaviors, but we were only able to assess teacher training and parental engagement in school policies as targeted interventions.

6.1. Mesosystem

Although the data did not allow for close examination of mesosystem factors such as school climate, specifically from the students’ perspective, the survey did address various types of programs found to be effective at improving climate such as prevention curriculum, social and emotional learning, behavioral modification, mentoring programs, programs promoting a sense of belonging for marginalized groups, etc., that were recoded into the school climate index. This variable was not found to be a significant predictor for bullying behaviors in our study, however, challenging the literature and failing to support our expectations. We suggest that this is mainly due to measurement limitations. While operationalizing this variable, we simply counted the number of school programs in place, but we were unable to consider the quality, duration, and intensity of these programs. The studies in the literature indicate that more intensive teacher and/or student programs that have more program elements, longer duration, and higher intensity are more effective in decreasing bullying (Ttofi & Farrington, 2011). Additionally, the role of reporting culture is another important factor to take into account. Research finds that positive school climate increases trust and a belief that concerns will be addressed effectively, and encourages students to seek help and report bullying and threats of violence (Eliot et al., 2010). Also, “[s]tudents are more willing to report misbehavior in schools with democratic authority structures and consistent enforcement of school rules” (Slocum et al., 2017, p. 123). Therefore, this increased transparency of reporting in schools with stronger safety/climate measures may underestimate the true protective effects of these measures in our study. Finally, our measure of school climate factors was dependent on the survey respondent’s role and vantage point, which may not fully capture this concept, as we discuss further below.

Several findings at this level were found to be significant and consistent with the literature, however, including the positive impact of teacher training on all three forms of peer harassment, and the risk factors of higher grade levels and greater enrollments on all three forms. Studies in the literature show that anti-violence teacher training increases teachers’ ability to identify and handle instances of conflicts, disputes, bullying, and harassment at schools (Chen et al., 2017). Researchers also find that teachers’ self-efficacy is negatively associated with the rates of school bullying and harassment, and teacher training increases teachers’ self-efficacy in preventing and intervening in such incidents (Fischer et al., 2021). The finding that the risk for these forms of peer harassment increases in higher grade levels is also consistent with the existing literature that emphasizes different developmental stages of children (Holmqvist Gattario & Lunde, 2023; Skoog et al., 2023).

Also at the mesosystem level, we found an additional protective factor for sexual harassment and bullying, but not cyberbullying, i.e., regular or random metal detector checks. Although not discussed in the current literature, this difference could reflect the fact that metal detector checks increase students’ sense of school surveillance that diminishes their engagement in these activities that are performed face-to-face, where cyberbullying occurs online and surveilled differently.

The presence of security cameras is also found to be a protective factor for sexual harassment, but not bullying or cyberbullying. Awareness of security cameras acts as a direct disincentive for individuals from engaging in sexual harassment, as video footage offers strong evidence for disciplinary action (Robertson et al., 2023). However, longitudinal studies also find that security cameras are not an effective deterrent for social disturbances, including bullying and racial/ethnic tensions, because such behaviors can remain undetected even if there are security cameras (Fisher et al., 2021). For example, cyberbullying and less visible forms of bullying (e.g., relational bullying and verbal bullying) cannot be captured by traditional surveillance systems, and students can continue engaging in these behaviors in the presence of security cameras.

Finally, we found mesosystem factors that served as risk factors. Threat reporting systems and sworn law enforcement officers at school increased the likelihood of cyberbullying but not sexual harassment or bullying. Although this is not discussed in the literature, it is possible that the presence of these practices contributes to students’ shift in bullying behaviors from face-to-face to online. Another explanation could be reverse causation, where the prevalence of cyberbullying has contributed to the greater use of these measures or that presence of these measures may increase reporting rather than prevalence, but we are not able to confirm this empirically.

6.2. Exosystem

Consistent with the literature, we found that neighborhood crime was a risk factor for all three forms of peer harassment (e.g., Schwartz et al., 2021). Exposure to and normalization of violence can contribute to students’ engagement in harassment behavior. Although not explicitly addressed in earlier studies, parental involvement in school policies, like positive parental relationships with young people discussed in the broader literature (Lereya et al., 2013; Doty et al., 2017; Kaltiala-Heino et al., 2016), was found to be a protective factor for all three forms. Parental involvement might also contribute to greater connection and care between parents and students, as suggested by positive youth development frameworks (D’Urso et al., 2021).

Challenging existing literature, we found that engagement with community-based mental health and social services was associated with higher prevalence rates of all three forms of bullying, whereas the involvement of community law enforcement/juvenile justice systems was associated with higher frequency of cyberbullying. It is important to note that our results indicate an association rather than a causal relationship. We suggest that this relationship might be a result of reverse causation, where higher rates of incidents are responded to by the school with outreach to community services for assistance. It may also be the case that the increased engagement of these services increases reporting rather than prevalence. Research has found that community involvement activities in schools reduce the number of discipline referrals, improve school climate, and increase access to mental health services (Sheldon & Epstein, 2002; Anderson-Butcher et al., 2022), which may potentially increase reporting and awareness of student misbehaviors.

6.3. Macrosystem

Finally, urbanicity was found to be a protective factor for cyberbullying. We find that schools in urban areas (city or suburban) are less likely to have frequent cyberbullying and more likely to have occasional and no cyberbullying compared to schools in non-urban (town or rural) areas. This finding is both consistent and inconsistent with the relevant literature which shows mixed findings (for example, see Zhu et al., 2021; Cabrera et al., 2024). Our findings may be due to more supervision and guidance about internet use, and higher digital awareness in urban settings compared to schools in non-urban areas. Also, students in non-urban areas may be more likely to spend time online due to fewer social interactions or extracurricular activities and this may increase their exposure to cyberbullying. As suggested by Cabrera et al. (2024), students in rural areas may experience more distress as a result of cyberbullying due to their smaller networks and limited exposure to additional friend groups and social outlets. Our hypothesis was that rural areas tend to encompass more conservative attitudes towards marginalized groups that could contribute to higher rates of bullying in these areas, but only an association with cyberbullying, and not with bullying or sexual harassment, was found.

6.4. Summary

Our findings about teacher training, parental involvement, neighborhood crime levels, school size, and school grade levels align with expectations. These findings suggest that both mesosystem and exosystem level factors play significant roles for predicting the frequency of bullying, sexual harassment, and cyberbullying. The macro level influence of urbanicity was found to be a significant predictor of cyberbullying as well. Overall, these findings support the socio-ecological framework and suggest that the likelihood of student misconduct such as bullying, sexual harassment, and cyberbullying is shaped by a broad context of interconnected interactions and influences across multiple levels of the socio-ecological model. Significant predictors were identified at each socio-ecological level, which demonstrate that factors across these levels collectively shape the frequency of these behaviors, and focusing on a single level may overlook important predictors.

Additionally, our findings reveal an interesting pattern about how ordered outcomes (no, occasional, monthly, and frequent) behave across the three types of incidents. This pattern becomes observable thanks to the analysis of generalized ordered logistic regressions. For bullying and cyberbullying, when the predictors increase (decrease) the probability of frequent and monthly incidents, it is accompanied by a corresponding decrease (increase) in the probability of mostly occasional incidents and somewhat no incidents. This pattern shows a reasonably linear progression from low to high categories (or vice versa), where contextual factors affect the presence of incidents but mostly the severity of these incidents. For sexual harassment, on the other hand, frequent, monthly, and occasional sexual harassment behave in a similar manner, while no sexual harassment moves in the opposite direction. When the predictors increase (decrease) the probability of frequent sexual harassment, they also increase (decrease) the probability of monthly and occasional sexual harassment, and these changes are accompanied by a corresponding decrease (increase) in the probability of no sexual harassment. This shows a more dichotomous pattern, where contextual factors affect whether sexual harassment happens or not.

The patterns of sexual harassment may differ from those of bullying and cyberbullying for several reasons. First, school policies and interventions may be more effective in reducing the occurrence of sexual harassment incidents, including occasional sexual harassment, whereas similar measures may primarily reduce the frequency rather than the occurrence of bullying and cyberbullying. Sexual harassment is regulated under federal law through Title IX, leading to stronger institutional accountability and deterrence mechanisms (U.S. Department of Education, 2020). In contrast, bullying and cyberbullying are not governed by a specific federal law but rather by state-level laws and local policies with varying investigation and prevention procedures (Cascardi et al., 2018). As a result, responses to bullying behaviors are often less formalized, and disciplinary measures tend to be less stringent compared to sexual harassment. This can sustain occasional bullying and cyberbullying even under policy interventions, while the same policy interventions can decrease occasional sexual harassment as well. Second, reporting sexual harassment tends to be more sensitive and tightly regulated, potentially leading to underreporting of occasional or less severe cases. Research finds that some school employees failed to report the incident to law enforcement, and fears of community and media response, and fear of retribution are some of the challenges of reporting sexual harassment (Grant et al., 2019). Due to these reporting issues, when sexual harassment occurs occasionally, schools may underreport these incidents, whereas they may continue reporting the bullying and cyberbullying cases.

7. Policy, Research, and Practice Recommendations

Our study suggests that school staff need to consider a comprehensive set of policies and practices that address risk and protective factors across the socio-ecological spectrum. Given limited resources, schools can first prioritize the common risk and protective factors, like teachers training, parental involvement, school grade level, school size, and neighborhood crime, that are associated with the prevalence of all these three behaviors. Teachers and administrators should be aware that these forms of harassing behaviors increase by grade level, school enrollment, and neighborhood crime and take steps to prevent and respond. Teachers should be trained to identify and respond to different forms of bullying, and parents should be engaged in schoolwide efforts to prevent these behaviors. The policy interventions should aim to decrease higher-frequency incidents as well as occasional incidents, particularly attention should be given to occasional bullying and cyberbullying. Reporting mechanisms across all forms of peer harassment should be standardized and strengthened to minimize underreporting, particularly sexual harassment.

The use of specific disciplinary practices yielded mixed results. Where some were associated with lower prevalence rates of harassing behaviors, others were associated with higher rates. As we have suggested, the direction of causality is unclear, and the impact of specific interventions on reporting could have affected the results, such that “risk factors” might better be conceptualized as factors that increase trust and reporting.

Future studies should include longitudinal data better suited to measuring causality so that the impact of specific practices and policies on prevalence rates is more clear. It is also important to capture not only the existence of school safety programs but also their duration and intensity, and quality. Such practices can influence students’ relationships with peers, teachers, and parents’, and their impact on positive youth development through mechanisms of connection and caring require improved measures and further investigation (D’Urso et al., 2021).

It is also crucial to have standardization in survey administration. The SSOCS survey allows for individuals in a variety of formal school roles to complete the survey (e.g., principal, teacher, counselor, resource officer), all of whom may have distinctive perspectives on the forms of harassment that are occurring, as well as the types of programs and other supports the school offers. We are not able to assess differences across these roles, however, since the survey does not capture those data. Perceptions of forms of harm measured by the survey are shaped by both the individual respondents’ vantage point and experiences, and the survey itself. For example, greater attention to the understanding, reporting and categorization of sexual harassment incidents (whether there is frequent or occasional sexual harassment) is needed to ensure accurate documentation.

Additional research is also needed to more accurately measure school climate, particularly from the perspective of students.

8. Limitations and Further Research

This study compares risk and protective factors for bullying, cyberbullying, and sexual harassment by identifying factors at the levels of mesosystem, exosystem, and macrosystem rather than solely individual, which has received the most attention in the literature. The study contributes to the literature by using a more recent nationally representative dataset, conducting an ordinal logistic regression rather than a binary logistic regression, and providing a comparison across different types of student misconduct. However, several limitations exist.

This study uses cross-sectional data without an experimental design. The findings of this study cannot be used to infer causal relationships. Further research can use longitudinal analysis or quasi-experimental research designs to measure the causality of relationships.

Second, the study uses a dataset that surveys school principals or other school staff and relies on the reported allegations of peer victimization. The actual bullying, cyberbullying, and sexual harassment may be different from the reported ones. Also, variations in survey respondents and their knowledge about the distinctive form of student misconduct may result in differences across schools. This measurement error in the dependent variable may increase the error variance and may cause a threat to the internal validity of the study. Furthermore, since the survey is filled out by school principals or school staff, the study could not include individual-level factors (such as age, race/ethnicity, gender) and microsystem-level factors (such as peer support, teacher support, and parent support) into its analysis. To address measurement error and missing variables, further research can test the empirical validation of socio-ecological model in the school context by using a different dataset that uses student responses concerning their victimization experiences (either officially reported or not reported).

It is important to recognize that the data for this study are based on perceptions of representatives of each responding school since there is no indication that they are based on official reports. In most cases, the individual completing the survey was the school’s principal. For example, in 2019–2020 school year data, 80% of surveys were completed by the school principal, followed by vice principal or disciplinarian (6.16%) (Kaatz et al., 2024). Also, 30% of survey respondents received help from other school personnel to complete the survey (Kaatz et al., 2024). We recognize that this raises many factors that could influence the validity of the data. For example, students may not always report or disclose their experiences and principals may be distanced from those individuals responding to such incidents. Principals may also prefer not to disclose negative behaviors that are occurring in their schools. We recognize these as limitations of the study, but we also argue for the usefulness of the study for several reasons. First, given the continued implementation of and support for this survey at the national level, it is one of the few comprehensive sources of data that we have on peer harassment in schools. Secondly, we argue that even though principals may not be aware of all the peer harassment that is happening in their schools, they are the primary decision maker when it comes to the implementation of policy and practice, so their perceptions matter and inform their support of various prevention and intervention strategies.

Additionally, some study variables, such as teacher training and school safety/climate are measured based on a number of program elements. Due to data unavailability, the duration and intensity of these elements are not taken into account. This may cause some limitations concerning the construct validity. Also, the public-use files do not share some school-level characteristics to preserve confidentiality, although these variables exist in the restricted-use file (such as FIPS state codes or county codes for schools, the percentage of female students, the percentage of students belonging to a racial/ethnic minority, the percentage of students eligible for free or reduced-price lunch, student-teacher ratio at schools). To address this limitation, further research may replicate this study by using the restricted-use file and accounting for those school-level characteristics, also further research may employ a multilevel modeling approach to capture higher-level policy or contextual influences. The harm done but peer harassment in schools requires us to continue to engage in research that informs policy and practice at every level to create safe spaces for all of our youth to thrive in schools.