Prevalence of and Factors Associated with School Bullying in Students with Autism Spectrum Disorder: A Cross-Cultural Meta-Analysis

Studies that meet all the eligibility criteria were included in this meta-analysis: 1) studies must include students diagnosed with ASD [autism, Asperger syndrome, pervasive developmental disorder-not otherwise specified (PDD-NOS)]; 2) studies must have information on prevalence estimates of school bullying, factors correlated with bullying, predictors of bullying, and/or outcomes of bullying; 3) study is designed as a cohort, cross-sectional, or case-control study, excluding qualitative, longitudinal, intervention, and theory studies; 4) studies are published in English, Japanese, or Korean; and 5) studies were published in peer reviewed journals by January 2018.

Electronic databases of social science and psychiatry literature were utilized for literature searches in 1) English (Academic Search Complete, CINAHL Plus with Full Text, Education Sources, E-Journals, ERIC, Medline, PsycARTICLES, PsychINFO, and SocINDEX with Full Text); 2) Japanese [Medical Online, CiNii, J-Stage, and Japan Medical Abstracts Society (Ichusi)]; and 3) Korean (DBpia, KISS, RISS, KYOBO Scholar, eArticle, and Newnonmun). Keywords for the literature search included all possible combinations of (a) autis* OR asperger OR pervas* developm* dis* OR PDD-NOS, (b) bull* OR victim* OR perpetrat* OR teas* or violen*, and (c) school* OR education* OR class* in the English databases; (a) autism OR ASD OR Asperger OR neurodevelopmental disorder OR pervasive developmental disorder, (b) ijime, and (c) school OR class in the Japanese databases; and (a) autism, Asperger, PDD-NOS and (b) peer-bullying, bullying, Wang-tta, school violence in the Korean databases. The database searches yielded 848 articles in English, 312 articles in Japanese, and 18 articles in Korean. Additionally, review of references in previous review/meta-analysis studies14, 28, 29, 30, 31, 32 yielded 78 articles in English and seven articles in Korean.

We reviewed the titles and abstracts of each study to determine eligibility for inclusion in this meta-analysis. This resulted in 62 articles in English, seven articles in Japanese, and nine articles in Korean. After a full-text review, 34 articles in English, one article in Japanese, and no articles in Korean remained (Fig. 1). The first author provided other authors with a manual of procedures for meta-analysis. The database searches and the literature review were conducted independently. In all steps, consensus decisions were made by two authors per each language, and discrepancies were resolved by discussion. Percentage agreement between the authors was 93% for English articles and 100% for Japanese and Korean studies. Supplementary Table 1 (only online)33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56 for a final list of the reviewed studies.

Fig. 1
Flow diagram illustrating search and selection of studies.

• Click for larger image
• Download as PowerPoint slide

The methodological quality of each study was assessed with the Appraisal Tool for Cross-Sectional Studies (AXIS).57 AXIS covers three domains: reporting quality (7 items), study design quality (7 items), and biases (selection, information, non-response; 6 items). Of these, the 13 items for study design quality and biases were used to assess the methodological quality of each study, and the seven items from reporting quality (how well methods, results, and discussion are described well) were excluded because they were irrelevant to methodological quality. AXIS only addresses selection, information, and non-response biases, but not confounding bias, a major threat to the study validity. Item 11 of the Critical Appraisal Checklist for Cross-Sectional Study58 was added to measure confounding bias. This item asks the question: “Could there be confounding factors that haven't been accounted for?”

Selection bias occurs when every unit in the sample frame does not have the same probability of being selected. Information bias occurs when systematic differences in the accuracy of data lead to a distortion in research results. Non-response bias occurs when the non-responders are systematically different from responders in the sample. Confounding bias occurs when an extraneous variable distorts the association between an independent variable and a dependent variable.

In the end, 14 items were used to assess the methodological quality of the included studies. All items were coded as 1 when a study satisfied a qualitative standard or as 0 when a study did not satisfy a qualitative standard or when assessment was not possible (unknown). Scores ranged from 0 to 14, with higher scores indicating better methodological quality. Two authors assessed cooperatively the methodological quality for each study and reached consensus decisions.

Information from each study was collected using an electronic data extraction sheet developed for this review. In each study, the following items were coded: publication year, culture (Western culture/Eastern culture), gender (percentage of boys in the sample), age (childhood: 5–12 years; adolescents: 13–22 years; mixed: 5–22 years), school settings (inclusive/ segregated/ mixed), measurement method (questionnaire/interview/mixed), bullying frequency criterion (dichotomous score: experience bullying yes or no/cut-off score: experience more than once a week, or twice a week, or once a month, or twice a month, or four times a year), assessment time-frame (month: last month or months/year: last or current year/life-time), and respondents (self/parent/teacher/peer). Age was coded based on both the age mean and the age range of the sample. Culture was classified as Western or Eastern culture if the study was conducted in the West or the East, respectively. Prevalence of bullying was coded in frequencies. Sample sizes and statistic values (χ², means and standard deviations, t value and p value, F value and p value, correlation coefficients, standardized β) describing relationships between associated factors and bullying experiences were also coded. Consensus coding was made by two authors for each language. When information on sample characteristics, or statistic values were missing or unclear, the first author contacted and requested information from the lead author of the studies. Overall, 13 authors were contacted, and two provided requested information.

Because the included studies varied significantly in their methods, analyses were conducted using a random effects model (Meta XL 5.3 and STATA 16.0), per the following procedure. Pooled prevalences, with 95% confidence intervals (95% CIs), of victimization, perpetration, and perpetration-victimization in general, verbal, physical, relational, and cyber forms were computed using the double arcsine transformation59 for students with ASD. When research participants responded ‘yes’ on any types of bullying questions or on a general question of being bullied by others or bullied others, it was categorized as “general” bullying. Their risk of bullying experiences was compared with that in typically developing students and students with other disabilities by calculating pooled odds ratios (ORs) with 95% CIs. Students with other disabilities were divided into nine groups, eight of which are based on the Individuals with Disabilities Education Act (IDEA) classifications60 as follows: hearing and visual impairment, emotional disturbance, intellectual disability, orthopedic impairment, other health impairment, specific learning disability, speech or language impairment, and traumatic brain injury. When type of disabilities was not available, they were assigned to the group of “various disabilities,” the ninth group. Weighted mean effect sizes (pooled effect sizes) and 95% CI were computed for degree of associations between associated factors and bullying experiences using Fisher's Zr transformations. Publication bias was assessed through visual inspection of Doi plots, as well as quantitative assessment of asymmetry through the LFK index.61 If the Doi plot indicated an existing asymmetry and the LFK index was >±1, the presence of publication bias was suggested, indicating a possible skewing of the pooled results. Heterogeneity was investigated using Q statistics, and moderation analyses were conducted to identify sources of heterogeneity.

Efforts were made to avoid violation of the independence of observations assumption62 as follows. If multiple studies were published from the same data set20, 23, 39, 53, 54 and had multiple statistical values with the same information, multiple studies were considered as a single study and the highest value was selected. If statistical values of a parent variable (e.g., internalizing) and nested variables (e.g., fear, depression) for an associated factor were presented in a study, only the value of the parent variable was included. When there were multiple statistical values of nested variables (e.g., fear, depression) for an associated factor in a study, their average effect size was used.

Meta-analyses based on very small number of studies are subject to problems in statistical synthesis.63 Generally, it is recommended to include five or more studies for data synthesis,63 six or more studies for a continuous moderator, and four studies per level of a categorical moderator.64 However, due to the limited number of studies available for the current meta-analyses, we set the threshold of a minimum of three studies for pooled prevalence estimates, pooled ORs, and pooled effect sizes; a minimum of three studies for analyses using continuous moderators; and a minimum of three studies per level of analyses using categorical moderators. These criteria are consistent with those used in previous meta-analyses of a limited number of studies.65, 66

Table 1 describes the results of pooled prevalence analyses. Supplementary Table 2 (only online) displays the forest plots and Doi plots of pooled prevalence estimates of bullying.

Table 1
Pooled Prevalence of Bullying and Test Results for Heterogeneity and Publication Bias

• Click for larger image
• Click for full table
• Download as Excel file

Pooled ORs for bullying in students with ASD were computed in comparison to typically developing students or students with other disabilities, including emotional disturbance, intellectual disability, other health problems, specific learning disability, and various disabilities (Table 2). Students with ASD faced a significantly higher risk for victimization than typically developing students (OR=2.35; 95% CI=1.57–3.53; p<0.001) and students with various disabilities (OR=1.98; 95% CI=1.30–3.02; p=0.001). On the other hand, their risk for victimization was not significantly different from students with emotional disturbance, intellectual disability, other health impairment, or specific learning disability. Pooled ORs for perpetration in students with ASD relative to both typically developing students and students with other health impairment were not significant.

Table 2
Risk of Victimization, Perpetration, and Perpetration-Victimization

• Click for larger image
• Click for full table
• Download as Excel file

Pooled effect sizes for the associations between factors and bullying were computed when meeting statistical synthesis criteria (≥3 studies). In result, eight factors for victimization (age, gender, ASD symptoms, deficits in social communication and social interaction, restricted, repetitive patterns of behavior, interests, or activities, externalizing symptoms, internalizing symptoms, inclusive school setting) and three factors for perpetration (gender, externalizing symptoms, internalizing symptoms) were included in this meta-analysis (Table 3). The forest plots and Doi plots are available in Supplementary Table 3 (only online).

Table 3
Pooled Effect Sizes and Test Results of Heterogeneity and Publication Bias

• Click for larger image
• Click for full table
• Download as Excel file

Victimization was positively associated with deficits in social interaction and communication, externalizing symptoms, and internalizing symptoms. Interestingly, a positive effect size was found for victimization and the inclusive school setting: this means that students with ASD who were integrated in the inclusive school setting experienced more severe victimization than those who were not. Perpetration was positively associated with externalizing symptoms. A significant heterogeneity between studies was noted for all pooled effect sizes, with an exception in two pooled effect sizes (gender and perpetration, internalizing symptoms and perpetration). Also, Doi plots and the LFK index suggested that most of the pooled effect sizes had publication bias (minor to major asymmetry, LFK=1.17–4.83), except for three (gender and victimization, gender and perpetration, restricted, repetitive patterns of behavior, interests, or activities and victimization; no asymmetry, LFK=0.08–0.53).

Maximum quality scores (14) were not achieved for any study included in the current meta-analysis. Most studies (97% ; see Item 2. Study design) used an appropriate study design for their purposes. A few studies reported the methods used to determine the sample size (11%; see Item 3. Sample Size Justification) and defined the population from which study subjects were selected (9%; see Item 5. Sampling Frame). Many studies (80%; see Item 8. Measurement Validity) used appropriate measurements. In most studies, the main findings were discussed in detail (71%; see Item 17. Justified Discussion), and there were no concerns identified for funding sources or conflicts of interest affecting study results (94%; see Item 19. Conflicts of Interest). Ethical consent from participants was reported to have been obtained in most studies (94%; see Item 20. Ethical Approvals).

In the assessment of bias, only a few studies (14%; see Item 6, Sample Selection) used randomly selected population samples, leaving a potential for sampling bias in the rest of the studies in this meta-analysis. However, non-responders were addressed and categorized (e.g., quantifying non-responders, using a statistical method to address missing data); numbers of non-responders were modest, minimizing potentials to introduce a bias; and information about non-responders was described in many studies (74%; see Item 7. Non-responders, 77%; see Item 13. Response Rate, 77%; see Item 14. Non-responders Information). In sum, the potential for non-responder bias was modest in the included studies. Half of the studies used measurements for which reliability had been verified previously (54%; see Item 9. Measurement Reliability), leaving the other half vulnerable to information bias. Confounding factors were considered and controlled in more than half of the studies (63%; see Item 11. Confounding), but not in some studies, leaving the potential for confounding bias. Finally, most studies provided internally consistent results (89%; see Item 15. Consistent Results), such that the risk of bias from inconsistent results was likely to be low (Supplementary Table 7, only online).33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56

Students with ASD face considerable difficulty in reciprocal social interactions, impairment in social communication skills,8 challenges in peer relationships,67, 68 and emotional-behavioral problems.69 These difficulties make them vulnerable to peer victimization.21, 50 This study sought to outline the bullying experiences of students with ASD and factors associated therewith through a meta-analysis of published studies in order to suggest guidelines for clinical care for students with ASD and to provide directions for future studies.

Pooled prevalence estimates of victimization, perpetration, and perpetration-victimization in students with ASD were 67%, 29%, and 14%, respectively, much higher than those for typically developing 6th–12th graders in the US (prevalences of 20%, 19%, and 7%, respectively).70 They are also higher than those for victimization (36%) and somewhat lower than those for perpetration (35%) in typically developing adolescents reported in a meta-analysis of 80 studies published in English.71 Risk for victimization in students with ASD was 2.4 times higher than that for typical developing students and two times higher than that for students with various disabilities. The risk for perpetration was not different from that of typical developing students and students with other health impairment. These findings suggest that students with ASD are at higher risks for experiencing bullying, in particular victimization, when compared to typically developing children and those with other disabilities.

Regarding victimization, students with ASD experienced the verbal form most frequently (58%), followed by relational (36%), physical (30%), and cyber-victimization (15%), suggesting that a special focus on prevention of and intervention in verbal and relational bullying leaving no traces is necessary in this population. Students with ASD were less likely to be exposed to cyber-victimization than other traditional forms of victimization. However, 15% is never a small figure. The prevalence of cyber-bullying for children with ASD will likely increase and become a greater focus in the future, as the number of students with ASD are increasingly using electronic devices and social networking services24, 72 due to their own interests and as a result of the COVID pandemic, which has forced increases in internet-based schooling and therapy services, as well as social isolation due to “sheltering-in-place.” These circumstances should lead to the prompt initiation of preventive measures and interventions for cyber-victimization.

In regards to perpetration, physical-perpetration was most common (27%), followed by verbal (26%) and relational (12%) forms in students with ASD. Given their core deficit in social communication, it is not surprising that physical and verbal forms of perpetration are more common than relational forms in students with ASD.73, 74 Some studies suggest that perpetration behaviors in students with ASD are related to comorbid conditions,16, 40 suggesting clinical care for comorbid conditions and teaching coping skills to manage negative feelings and frustration are beneficial for preventing and/or reducing perpetration behaviors in students with ASD.

Social difficulties in students with ASD were positively related to bullying experience. Difficulties in forming positive peer relationships, as well as the lack of skills with which to cope with peer conflicts, might contribute to increased risks for bullying experiences.75, 76 Comorbid conditions in students with ASD also are correlated with risks for being involved in bullying experiences. Specifically, externalizing symptoms are significantly correlated with victimization and perpetration, whereas internalizing symptoms are significantly correlated with victimization. The disruptive nature of these comorbid conditions may cause peer rejection and retaliation, as well as convey heightened vulnerabilities, in students with ASD, leading to an increased risk for victimization.77 Additionally, emotional outbursts or aggressive behavior, symptoms of externalizing behaviors, may be considered as perpetration behaviors.78

While comorbid conditions in children with ASD increase the risk for bullying experiences, it has also been established that bullying is related to later development of psychopathology.79, 80 One study reported that 45% of adults with Asperger's syndrome had long-term sequelae from prior bullying experiences.81 Another study reported increased levels of anxiety in individuals with ASD who experience bullying victimization.22 A vicious cycle between bullying and psychosocial problems exists.82 Therefore, careful attention must be given to the children with ASD in order to protect these already vulnerable children from long-term consequences of bullying.

Inclusive school settings are significantly correlated with increases in victimization for students with ASD. Bullying occurs in the presence of a power imbalance between bullies and victims.83 Inherent power imbalance exists between students with ASD and typically developing peers due to impairments in social communication, social skills, and comorbid conditions in students with ASD. Our findings indicate that simply exposing students with ASD to inclusive school settings, without appropriate preparation, not only fails to promote acquisition of social skills and development of peer relationships for students with ASD, but also it poses harm to students with ASD by increasing their risks for peer victimization. For successful outcomes in inclusive education for students with ASD, as well as typically developing children, ongoing efforts are required for peer-mediated interventions against bullying;84 disability awareness education for students, families and school staff;85 supervision during recess time; and firm and consistent disciplinary methods for bullying across various settings, including homes, schools, and communities.86

Consistent with previous systematic review articles,14, 31, 32 we also identified significant heterogeneity within the included studies. Such heterogeneity may distort estimates of pooled prevalence and effect sizes in a meta-analysis. In order to mitigate the impact of heterogeneity, we used a random effects model in analyses to allow for harmonization of heterogenity of results between studies.64 We then conducted moderation analyses to identify sources of heterogeneity. The moderation analyses showed that the correlation between internalizing symptoms and perpetration is weaker in more recent studies. This may be due to differences in study methods over the time. The prevalence of general bullying was significantly higher in Eastern rather than Western culture. Eastern culture is characterized as collectivistic, with Western culture being more individualistic. In comparison to individualist culture, interdependency, cohesion, and unity are regarded as important values of the group in collective cultures.87 As such, the desire to follow collective norms to ensure harmony is the major force regulating behavior in Eastern cultures.88 Thus, it is plausible that peers in collective cultures may have rejected and bullied students with ASD who are different from them. In addition, inclusive education has been adopted in Eastern countries later than in Western countries, thus less experienced teachers, school personnel, and policymakers with respect to inclusive education might face more challenges in bullying prevention and intervention.

Victimization in verbal and relational bullying types was more frequent in younger students with ASD (5–12 years) than in adolescents with ASD (13–22 years). This pattern is consistent with that in typically developing students. For example, victimization was more frequent in 6th grade (approximately 12 years old) than in 7th–10th grade students (approximately 13–16 years) in a survey of 6th–10th graders in schools from all areas of the US.89 A Korean study also reported that victimization in students with other disabilities was more frequent in elementary schools (approximately 8–12 years) than in high schools (approximately 16–18 years).90 Fewer comorbid conditions (e.g., internalizing and externalizing symptoms) in older children with ASD than in younger children with ASD,91 along with more coping strategies developed with longer educational and social learning experiences,92 might have resulted in such an age pattern.

Students with ASD were significantly more likely to be targets of physical-victimization when schooled in the mixed school setting (inclusive setting+segregated setting) than in the inclusive school setting. Since sophisticated social communication skills are necessary to use relational aggression successfully,73, 74 students with other disabilities included in segregated school settings may use physical aggression more frequently than relational aggression towards students with ASD due to their deficits in social communication skills.

Of the study methods, no moderators (bullying measurement method, criteria for prevalence estimates, assessment time-frame, respondents) had a significant effect on bullying prevalence. However, methodological quality had a significant impact on the prevalence of cyber-victimization and relationships between gender and perpetration and between internalizing and perpetration. Additionally, sampling problems (i.e., sample size justification, representativeness of the study subjects for target population) were common in most of the studies. Future studies need to make efforts to improve the methodological quality of their research and to minimize the possibility of bias stemming from sampling processes.

As in prior meta-analytic studies.14, 65, 93 Doi plots and LFK indices indicated publication bias in some estimates of pooled prevalence and effect sizes in our study. While this suggests that publication bias may have skewed the pooled estimates, it is difficult to adjust for publication bias in computing estimates of pooled prevalence and pooled effect sizes in the presence of between-study heterogeneity.94, 95 Independent replications of our findings are warranted when more studies are accumulated and when methods for dealing with publication bias are advanced.

The limitations of this study include the following: 1) The small number of studies included in the current meta-analysis did not allow for the examination of the impact of more moderators and factors associated with bullying experiences. 2) The limited number of studies conducted in Eastern cultures makes it challenging to fully explore the impact of cultural differences on bullying in our study population. 3) Heterogeneity and publication bias were identified and may affect some estimates of pooled prevalence and effects sizes. Notwithstanding, our findings indicate that students with ASD are at higher risks for victimization and perpetration of bullying. The severity of victimization was higher when students with ASD were younger and were in inclusion classroom settings. Special attention is warranted to assist and/or promote adjustment of young students with ASD when they are placed in inclusion classroom settings. Clinicians, parents, teachers, and school personnel (including principles and school nurses) must be attentive when students with ASD present with high levels of social difficulties and internalizing and/or externalizing symptoms. This should lead to assessments for whether the students with ASD are experiencing bullying. These symptoms are risk factors requiring increased attention in order to protect students with ASD from bullying and its detrimental consequences. Universal interventions for both students with ASD and other members of the school community, including classmates, teachers, and school personnel are crucial. Clear rules and expectations, as well as specific intervention protocols, about bullying should be established for all students, teachers, and school personnel. Bullying prevention education should be provided regularly for all. Further studies are warranted to identify other factors associated with bullying experiences in students with ASD and what prevention and/or intervention practices are effective in reducing this all-too-common form of school violence worldwide.

Supplementary Table 1

Characteristics of the Reviewed Studies

Supplementary Table 2

Forest Plots and Doi Plots of Pooled Prevalence

Supplementary Table 3

Forest Plots and Doi Plots of Pooled Effect Sizes

Supplementary Table 4

Forest Plots for Moderation Analyses in Pooled Prevalence for General and Verbal Victimization

Supplementary Table 5

Forest Plots for Moderation Analyses in Pooled Prevalence for Physical and Relational Victimization

Supplementary Table 6

Forest Plots for Moderation Analyses in Pooled Prevalence for General Perpetration

Supplementary Table 7

The Result of Methodological Quality Assessment in Reviewed Studies