Effectiveness of school-based programs to reduce bullying perpetration and victimization: An updated systematic...

Bullying first emerged as an important topic of research in the 1980s in Norway (Olweus), and a recent meta-analysis shows that these forms of aggression remain prevalent among young people globally (Modecki et al.). Prominent researchers in the field have defined bullying as any aggressive behavior that incorporates three key elements, namely: (1) an intention to harm, (2) repetitive in nature, and (3) a clear power imbalance between perpetrator and victim (Centers for Disease Control and Prevention; Farrington). There are many negative outcomes associated with bullying perpetration, such as: suicidal ideation (Holt et al.), weapon carrying (Valdebenito et al.), drug use (Ttofi et al.), and violence and offending in later life (Ttofi et al.). Bullying victimization too is associated with negative outcomes such as: suicidal ideation (Holt et al.), anxiety, low self-esteem and loneliness (Hawker& Boulton). Therefore, school bullying is an important target for effective intervention, and should be considered a matter of public health concern.

The objective of this review is to establish whether or not existing school-based antibullying programs are effective in reducing school-bullyng behaviors. This report also updates a previous meta-analysis conducted by Farrington and Ttofi. This earlier review found that antibullying programs are effective in reducing bullying perpetration and victimization and a primary objective of the current report is to update the earlier analysis of 53 evaluations by conducting new searches for evaluations conducted and published since 2009.

Systematic searches were conducted using Boolean combinations of the following keywords: bully*; victim*; bully-victim; school; intervention; prevention; program*; evaluation; effect*; and anti-bullying. Searches were conducted on several online databases including, Web of Science, PscyhINFO, EMBASE, EMBASE, DARE, ERIC, Google Scholar, and Scopus. Databases of unpublished reports, such as masters' and doctoral theses (e.g., Proquest) were also searched.

Results from systematic searches were screened thoroughly against the following inclusion criteria. To be included in this review, a study must have: (1) described an evaluation of a school-based antibullying program implemented with school-age participants; (2) utilized an operational definition of school-bullying that coincides with existing definitions; (3) measured school-bullying perpetration and/or victimization using quantitative measures, such as, self-, peer-, or teacher-report questionnaires; and (4) used an experimental or quasi-experimental design, with one group receiving the intervention and another not receiving the intervention.

Of the 19,877 search results, 474 were retained for further screening. The majority of these were excluded, and after multiple waves of screening, 100 evaluations were included in our meta-analysis. A total of 103 independent effect sizes were estimated and each effect size was corrected for the impact of including clusters in evaluation designs. Included evaluations were conducted using both randomized (n = 45; i.e., randomized controlled trials/RCTs) and nonrandomized (n = 44; i.e., quasi-experimental designs with before/after measures; BA/EC) methodologies. All of these studies included measures of bullying outcomes before and after implementation of an intervention. The remaining 14 effect sizes were estimated from evaluations that used age cohort designs. Two models of meta-analysis are used to report results in our report. All mean effects computed are presented using both the multivariance adjustment model (MVA) and random effects model (RE). The MVA model assigns weights to primary studies in direct proportion to study level sampling error as with the fixed effects model but adjusts the meta-analytic standard error and confidence intervals for study heterogeneity. The RE model incorporates between-study heterogeneity into the formula for assigning weights to primary studies. The differences and strengths/limitations of both approaches are discussed in the context of the present data.

Our meta-analysis identified that bullying programs significantly reduce bullying perpetration (RE: odds ratio [OR] = 1.309; 95% confidence interval [CI]: 1.24-1.38; z = 9.88; p < .001) and bullying victimization (RE: OR = 1.244; 95% CI: 1.19-1.31; z = 8.92; p < .001), under a random effects model of meta-analysis. Mean effects were similar across both models of meta-analysis for bullying perpetration (i.e., MVA: OR = 1,324; 95% CI: 1.27-1.38; z = 13.4; p < .001) and bullying victimization (i.e., MVA: OR = 1.248; 95% CI: 1.21-1.29; z = 12.06; p < .001). Under both computational models, primary studies were more effective in reducing bullying perpetration than victimization overall. Effect sizes varied across studies, with significant heterogeneity between studies for both bullying perpetration (Q = 323.392; df = 85; p < .001; I  = 73.716) and bullying victimization (Q = 387.255; df = 87; p < .001; I  = 77.534) outcomes. Analyses suggest that publication bias is unlikely. Between-study heterogeneity was expected, given the large number of studies included, and thus, the number of different programs, methods, measures and samples used.

We conclude that overall, school-based antibullying programs are effective in reducing bullying perpetration and bullying victimization, although effect sizes are modest. The impact of evaluation methodology on effect size appears to be weak and does not adequately explain the significant heterogeneity between primary studies. Moreover, the issue of the under-/over-estimation of the true treatment effect by different experimental designs and use of self-reported measures is reviewed. The potential explanations for this are discussed, along with recommendations for future primary evaluations. Avenues for future research are discussed, including the need further explain differences across programs by correlating individual effect sizes with varying program components and varying methodological elements available across these 100 evaluations. Initial findings in the variability of effect sizes across different methodological moderators provide some understanding on the issue of heterogeneity, but future analyses based on further moderator variables are needed.