The ACT Out! Ensemble was founded in 1995 and is currently operated by Claude McNeal Productions (CMP), a professional theater troupe incorporated as a not-for-profit [1]. The ensemble uses scripted content (scenarios) generated to meet an audience’s needs and transforms it into improvisational, interactive theater performances. The actors deliver performances focused on a variety of topics salient to youth and adolescents, including bullying, diversity, inclusion, and substance use [1]. A facilitator pauses the action at the end of each scenario and enables discussion between attendees and the actors, who remain in character for the discussion [1]. For example, a scenario about bullying may instruct the group to create a scene with “an example of a male student making a female student physically uncomfortable.” After the actors present an interpretation of that scene, the facilitator pauses the action and asks the audience “whether the male character meant to make the female character uncomfortable,” with the actors participating, in character, in the audience discussion.

The premise behind the ACT Out! Social Issue Theater intervention, outlined in the trial protocol, is that dramatic performances can enable emotional catharsis, thereby allowing new ways of feeling and thinking about behaviors and attitudes [2]. In other words, there is likely a difference between students discussing or attending a lecture on an issue like bullying in their own lives (where their own identity has weight and affects perceptions) and students’ emotional responses to a scenario that seems real and familiar (eg, bullying) but that is occurring with characters rather than with themselves or their peers. The latter case would theoretically enable students to process their reactions to bullying separately from their own or their peers’ identities. This is facilitated by the high caliber of talent involved in the intervention performances; shows by CMP have received positive reviews from, among other venues, the New York Times, NBC, and Time magazine [3]. For this trial, CMP developed and revised 15 vignettes (5 per participating grade level) addressing bullying and cyberbullying using principles of social-emotional learning (SEL) and reviewed the content with the research team. A presentation of the 5 psychodramatic vignettes was planned to last approximately 1 hour, including student interaction with the characters, and these performances constituted the intervention for this trial. This paper describes a rigorous evaluation, through a cluster randomized controlled trial, of the ACT Out! Ensemble’s theater performance, addressing bullying and SEL.

In the United States and internationally, schools, school-based professionals, and policymakers have begun focusing on positive development models as a means of addressing the numerous, complex, and detrimental behavioral patterns and associated outcomes (eg, bullying, mental health problems, self-harm, and substance use) observed among youth and adolescents [4]. Such approaches deliberately avoid a deficit approach (“fixing what is wrong”) and emphasize the development of assets or protective factors in youth. Among the most common and conceptually similar positive development models are positive youth development [5], which emphasizes skills development, healthy relationship development, supportive community systems, and SEL. SEL focuses on the instruction of skills such as social problem-solving, recognizing emotions in others, and emotional self-regulation [6]. There have been numerous SEL programs implemented and evaluated in schools in recent years; a summary of over 300 studies contained in 4 meta-analyses identified generally positive short-term outcomes [7] across multiple domains (eg, substance use). In general, performances by the ACT Out! Ensemble are structured to model aspects of SEL, such as healthy relationships, regardless of the additional topic being addressed (eg, bullying). We have summarized additional content related to SEL and social-emotional competence (SEC) as it pertains to this study in our published protocol [2].

School bullying is frequently mentioned among the detrimental behaviors addressed by SEL programs [8,9]. Bullying is an unfortunate reality for youth attending US schools; a meta-analysis of 80 studies (youth aged 12 to 18 years) found a 35% student-level prevalence of traditional bullying and a 15% prevalence of cyberbullying [10]. Being bullied in childhood and adolescence has been associated with long-term, negative consequences that persist into midlife in areas such as mental and general physical health and lower socioeconomic status attainment [11]. Further, bullying victimization appears likely to cause notable increases in anxiety and depression among those victimized [12].

A recent meta-analysis of traditional bullying identified 65 school-based bullying prevention programs, but only 8 had been evaluated more than once [13]. In general, such programs tend to be slightly more effective in reducing bullying perpetration and less effective in reducing victimization [14]. Mean values from meta-analyses have been somewhat consistent in terms of victimization, reporting reductions of 15-16% [14] and 17-20% [15]. A separate meta-analysis, focused only on cyberbullying, reported a mean reduction in victimization of 14-15% [16]. However, a meta-analysis of victimization studies limited to randomized controlled trials with a high level of rigor reported a rather small effect size (standard mean difference of -.09) [17]. Each of the cited meta-analyses note high levels of heterogeneity in the types of programs and outcomes across included programs and studies; importantly, these differences extended to study design and rigor, with some scholars noting that beneficial effects appear to be weaker when measured as part of randomized controlled trials [18].

As one might expect, reductions in victimization tend to be larger for more intensive and multifaceted programs, but implementing such programs can be both expensive and complicated [11,19,20]. For example, a summary of nontargeted (general population), relatively efficacious bullying prevention programs for US elementary schools found that program durations ranged from 11 weeks to 3 years [21]. Even efficacious programs described as “brief” prevention curricula can last 1 week or more and involve multiple interlocking components [22]. Further, several programs that have reported favorable results in efficacy trials have not always produced the same results in effectiveness (“real world”) trials, potentially due to issues with implementation fidelity and existing, confounding antibullying programming [23,24]. In concluding their report from a recent effectiveness trial, Rapee et al [24] noted, “clearly, producing a sizeable impact on school-based victimization is extremely difficult.” Therefore, there is a demonstrated need for inexpensive, simple-to-implement bullying prevention programming, but achieving positive outcomes from such interventions is likely to be especially challenging. For this reason, we believe that innovative or out-of-the-box strategies to address bullying merit serious consideration.

Given the difficulty in addressing school-based bullying with lengthy and multipartite curricula, one might wonder why a short (1-hour) dramatic performance would be hypothesized to have even a short-term effect on SEC or bullying. A small body of literature has examined psychodrama as a prevention or behavior-change mechanism in youth, but these studies have covered diverse behaviors [25], have involved multiple, separate components such as teacher training [26], or have used students or school employees rather than professional actors as dramatis personae [27]. ACT Out! Social Issue Theater is different than each of these examples because it uses trained, professional actors and requires no involvement from schools outside of planning the visit (when implemented outside of a study). We were unable to find a precedent for this intervention structure in the literature.

Our decision to analyze this intervention was based on our a priori understanding of the value this brief intervention might yield as well as the remarkable community- and school-level support for the program. Prior to this study, more than 500,000 individuals had viewed a performance by the ACT Out! Ensemble [1], providing a notable depth of informal, qualitative evidence supporting the program. Uncontrolled evaluations of the program from 2015 also suggested substantive behavioral benefits [1]. Thus, given the importance of both SEL and bullying prevention in schools, and the unique position occupied by the ACT Out! Ensemble, we determined that an independently conducted, randomized controlled trial of this intervention was a valuable contribution to the prevention literature.

This study primarily aims to assess whether a 1-hour exposure to ACT Out! Social Issue Theater is superior to treatment as usual for developing SEC and reducing bullying (both bullying behavior and victimization) in elementary, middle, and high school students at a 2-week posttest. Secondarily, the study aims to determine whether the same intervention is superior to treatment as usual in developing specific subdomains of SEC (social awareness, emotion regulation, relationship skills, and responsible decision making) among middle and high school students at a 2-week posttest. Finally, the study also aims to assess student receptivity to ACT Out! Social Issue Theater using previously validated measures indicating student agreement with positive (eg, “enjoyable”) and negative (eg, “boring”) adjectives. Additional details are available in the trial protocol [2]. All outcomes were measured at the individual participant level, but randomization occurred at the cluster level (classroom) because performances are intended to be delivered to groups and because research literature [23,24] has indicated that pre-existing school-level programs addressing bullying and SEC often vary between schools and may contribute to statistical noise in randomized trials using school as the cluster (eg, treatment as usual may not be consistent between schools).

The ACT Out! trial was a proof-of-concept cluster randomized superiority trial with 2 groups and 1:1 allocation. The unit of measurement was individual students, but the unit of randomization was the classroom, stratified by school (with 1 exception, Multimedia Appendix 1). For each school, half of the classrooms were randomly assigned to the intervention arm and half to the control arm. Schools with an odd number of classrooms had a single classroom randomly selected for exclusion (though if the school requested, that classroom was permitted to complete the survey for appearance’s sake, and the results were then discarded by the study team).

The ACT Out! trial was conducted among 12 public and charter schools in Indiana: 4 schools in Marion County, 3 in Ripley County, 2 in Boone County, 2 in Lawrence County, and 1 in Monroe County. For reasons described in the protocol, clusters were selected only from grades 4, 7, and 10 [2]. All students in the selected classrooms and schools were eligible to participate. As planned, we recruited schools until meeting a threshold of approximately 80 participating classrooms (around 1594 students) across both conditions.

Schools were selected based on their willingness to participate in the project as described, which included classroom-level randomization and inclusion of all eligible classrooms in the study’s allocation processes. Authorizing officials for schools or school corporations were required to provide a signed letter of agreement prior to participating in the study. At the individual level, the project used a waiver of parental consent (opt out), as approved by the institutional review board. Parents and legal guardians were permitted to review study procedures and were provided with a description of the study a minimum of 2 weeks prior to any individual-level interaction with subjects, along with instructions for how to opt out; students, their parents, and their guardians all had the ability to opt a student out from participating either formally or by survey noncompletion. The rationale for this approach was a combination of the low risk posed by the study as well as the desire to avoid unintentional exclusion of underrepresented minorities and high-risk populations, as described in the protocol [2]. This study and all consent procedures were carried out according to, and approved by, the Indiana University Institutional Review Board.

All measured outcomes were prespecified in the clinical trial registration and the published protocol [2], along with the rationale for their selection, and have been validated. Unfortunately, certain outcomes were not possible to collect due to the COVID-19 pandemic, which led to mandated school closures in the state during part of the data collection period (Multimedia Appendix 1). However, we do not have reason to suspect that the data we collected prior to closures were substantively affected.

Thus, this study collected the following 2 primary outcomes: (1) social-emotional competence and (2) bullying behavior and bullying victimization. Overall social-emotional competence was measured at baseline and at 2-week posttest using the Delaware Social-Emotional Competency Scale (DSECS-S) [28]. This scale prompted students to “Please read each statement and mark the response that best shows how much it is like you,” with response options of 1=Not like me at all, 2=Not much like me, 3=Somewhat like me, and 4=Very much like me. An example statement from the scale is, “I can control how I behave.” This scale demonstrated good internal consistency [intention-to-treat (ITT) pretest α=.78] for the study sample. Bullying behavior and experiences of being bullied (victimization) were measured at baseline and at 2-week posttest using the Bullying and Cyberbullying Scale for Adolescents (BCS-A) [29]. These questions measured the number of times (between 0 and 4+) in the past 2 weeks that students “bullied another school student” or, separately, “had been bullied.” These sections were further separated into subsections for “online/on the internet or mobile phones” (eg, cyberbullying behaviors) and “offline/face-to-face” (including physical, verbal, and relational behaviors).

Further, the study collected the following 2 secondary outcomes: (1) receptiveness to the intervention and (2) prespecified subanalyses of social-emotional competence for seventh- and tenth-grade students. Student receptivity to the intervention \was measured at 2-week posttest (intervention arm only) using questions to assess the degree to which they found the intervention to be enjoyable, interesting, a waste of time, boring, understandable, difficult to understand, believable, important, and helpful [30]. Social-emotional competence subdomains (social awareness, emotion regulation, relationship skills, and responsible decision-making) were measured at baseline and 2-week posttest (seventh and tenth grades only) using scales from the Washoe County School District Social-Emotional Competency Assessment (WCSD-SECA) [31]. These scales prompted students to “Please tell us how easy or difficult each of the following are for you,” with response options of 1=Very difficult, 2=Difficult, 3=Easy, and 4=Very easy. An example item from one scale is, “Getting along with my teachers.” Though the scales were previously developed to be reliable and valid [31], the items are relatively heterogeneous, perhaps contributing to their rather mediocre internal consistency with this sample [ITT pretest α=.57 (social awareness), .65 (emotion regulation), .68 (relationship skills), and .66 (responsible decision-making)].

The rationale for choices made in preparing the sample size calculation is provided in the protocol [2]. We estimated the sample size required to detect a moderate effect (Cohen d=0.30) with a 2-sided significance of .05 and a power of .80 to be 340 participants. We estimated an intraclass correlation of 0.153 based on a prior school-based cluster study on cigarette smoking with a similar methodology [2] and assumed approximately 20 students per classroom, yielding a design effect of 3.907. We took the resultant estimate of 1328 students and multiplied it by 1.2 to account for attrition and potential loss of matched pairs due to survey matching procedures, producing the final sample size target of 1594 students across approximately 80 classrooms.

This study required an anonymous procedure to match students’ surveys between pretest and posttest while remaining compliant with the requirements of the institutional review board. As described in the protocol [2], even recent meta-analyses had not identified a best-practice solution to such a dilemma [33]. Thus, the project team developed and used a novel anonymous matching procedure based on unique self-generated identification code elements and machine-assisted weighted matching. This approach was sufficiently complex that it required a separate full-length manuscript to articulate [34], and a complete description would extend well beyond the scope of this paper.

The formal study start date was October 16, 2019, and interventions were delivered from November 6, 2019, to February 28, 2020. School recruitment was terminated in February once the anticipated numbers of clusters and participants reached the planned total. Initially, a total of 13 schools from across Indiana participated in the trial, comprising 84 classrooms for the eligible grade levels. Of those 84 classrooms, 42 were randomized to the intervention arm and 42 were randomized to the control arm. One school did not follow protocol and failed to correctly administer pretest surveys to either arm prior to intervention delivery, so it (4 classrooms) was summarily removed from the trial. In addition, 3 schools provided incorrect counts of classrooms to be randomized (3 control classrooms and 1 intervention classroom), so sequences were generated that included classrooms that did not exist. Upon discovery of this discrepancy, sequences and assignments were not altered because it would have affected allocation concealment. Thus, despite 1:1 allocation, the number of baseline classrooms was 76 (37 control and 39 intervention classrooms). Finally, an additional 11 classrooms (5 control and 6 intervention classrooms) at a single school were slightly delayed in completing posttests, and then schools were shut down for the academic year due to COVID-19 prior to data collection. Thus, the number of classrooms that completed posttests was 65 (32 control and 33 intervention). Table 1 shows the baseline characteristics of the 2 trial arms for all students who provided data, excluding blank surveys.

Interrater reliability for fidelity coding was excellent (96.5% concordance). Fidelity to the intervention was high for all grades, though some seventh- and tenth-grade classrooms were not able to proceed through all 5 scenarios due to timing constraints. These instances were noted distinctly from fidelity because they resulted from longer-than-expected time completing the pretest survey immediately prior to the intervention, a factor that would not exist outside of the study. Fidelity was computed as the sum of all completed checkpoints for all clusters divided by the sum of all possible checkpoints for all clusters, within each grade. For the 5 fourth-grade classrooms, intervention fidelity was 100%. For the 15 seventh-grade classrooms, fidelity was 96.9%, but the fifth scenario was excluded for all clusters due to time constraints. For the 14 tenth-grade classrooms, fidelity was 98.4%, but 3 checkpoints for 1 classroom were excluded (fidelity checker was meeting with administrators), the third scenario was excluded for 2 clusters, the fourth scenario was excluded for 3 clusters, the fifth scenario was excluded for 3 clusters, and both the fourth and fifth scenarios were excluded for 1 cluster, all due to time constraints.

The ACT Out! Social Issue Theater trial was a prespecified and preregistered cluster randomized controlled trial conducted by a research team external to the program developers and supported by additional analysts who both were independent of the research team and the developers. Study findings were mixed.

For school-based programs, implementation fidelity tends to be inconsistently documented and highly variable [39]. There is some evidence that achieving good fidelity may require intensive training of those delivering the intervention (eg, teachers, [40]) and may depend on school- and teacher-level variables [41]. Since the ACT Out! intervention was an improvisational and interactive intervention by actors, there was little precedent as to whether implementation fidelity would be achievable. We found that when provided with guidelines for core content elements, the professional actors in CMP were able to deliver nearly all (96.9% to 100%) prespecified content (Multimedia Appendices 3-5), even accounting for variance in student responses and different actors playing different roles, distinguishing this intervention from many school-based interventions. However, ACT Out! performances were short in duration and the actors had the delivery of the performance as their primary purpose; in contrast, teachers and schools must balance many different requirements simultaneously, involving the delivery of multiple sessions, and so high fidelity is conceptually reasonable to expect and may be a benefit of programs of this type.

The degree to which students report enjoying a program or finding it to be realistic and engaging can be interpreted as an indicator of program quality and is a common component of process evaluation [30]. The ACT Out! intervention was very positively received by students across all specified metrics. Although not an indicator of emotional competence, we infer that this may be interpreted as a partial measure of emotional response to the intervention.

One of our primary hypotheses was that the ACT Out! intervention would improve students’ SEC in the short-term; this was suspected to be the mechanism through which the proposed emotional catharsis of psychodrama [2] could be measured. This study did not support that hypothesis, with analyses demonstrating neither statistical nor clinical significance, except improvement in emotion regulation regardless of treatment condition for seventh- and tenth-grade students.

During the project kickoff meeting, individuals at CMP expressed concern about quantitatively measuring SEC, at one point asking the research team, “How do you measure the sunrise?” We collected SEC data using 2 different tools that approached SEC in complementary ways [28,31]; however, optimal measurement of SEC remains a topic of debate, even among the national SEL workgroup [42]. Thus, we note (as with all such measurement) that the trial did not definitively find that the ACT Out! intervention had not engendered SEC development; rather, it found that SEC, as measured by the DSECS-S and WCSD-SECA (2 validated tools), was not affected by the intervention. It does not necessarily follow that those tools are the optimal or correct ways to measure students’ responses to the ACT Out! intervention (indeed, WCSD-SECA subscore reliability for this sample was suboptimal). It may also be the case that there was a partial ceiling effect [43] on overall SEC, as the baseline scores were relatively high: 3.202 and 3.176 for control and intervention groups, respectively, on a scale from 1-4, potentially making improvement from any source more difficult to achieve and measure. Given that all schools in the state already are required to offer SEL programming, this may also suggest that an intervention such as this would more appropriately be tested with subgroups of individuals who have lower baseline SEC (eg, not as a universal SEL program, but as an indicated program).

We hypothesized that the classrooms viewing the ACT Out! intervention would report reduced bullying victimization and perpetration [both traditional (physical, verbal, relational) bullying and cyberbullying] relative to the control classrooms. There was little evidence for an effect on perpetration, though 1 of the 4 models indicated the potential for slight increases for physical bullying in the intervention arm. There was also some evidence of a small reduction in cyberbullying victimization attributable to the intervention, though not for the traditional forms of bullying. That cyberbullying might be influenced separately from traditional bullying is reasonable, as cyberbullying victimization is unique in several ways, such as where it occurs (in a digital space “outside” of school) and its ubiquity [44]. Further, some of the dramatic scenarios (2 of 5 for both seventh and tenth grades) explicitly focused on cyberbullying as opposed to traditional bullying.

There was also evidence that physical bullying victimization was lower among the control group than the intervention group at posttest, though this was not an iatrogenic effect since neither group reported increased victimization scores. This finding may have been related to the issue we discuss subsequently.

Students reported clinically and statistically significant reductions in physical, verbal, and relational bullying victimization at posttest relative to pretest in aggregate (pooled across both study arms). This does not directly address the study hypotheses, as improvements were seen in students who did not participate in the intervention. However, this study examined students across 3 grade levels (fourth, seventh, and tenth), with interventions offered at different times over the course of nearly 6 months, in 12 schools and 65 classrooms, with a relatively short timeframe between pretest and posttest (14 to 27 days). Thus, it was implausible (though not impossible) that an external factor outside of the study was responsible for this finding.

We therefore considered effects that may have resulted from the study procedures. It was possible that participants’ responses at posttest were affected by completing the same items at pretest, either due to guessing and wanting to affect the study hypotheses or simple item-related biases introduced by familiarity with the questions [45]. However, some evidence suggests that questionnaire items do not affect student behavior [46]. Further, if this were the case, we would expect to have seen similar effects for cyberbullying victimization, bullying perpetration, and SEC, which we did not. We were also hesitant to ascribe these findings to regression to the mean [47]. The decrease was observed only for a specific subset of variables; as was already noted, bullying is common in school-based settings, so the likelihood that a large group of students from disparate settings and grades would significantly deviate from the population mean for bullying frequency is not conceptually strong. In addition, the impact of outlier cases on bullying frequency was minimized by the design, since the instance count in the questionnaire terminates at 4 (“4 or more times”). We also carefully avoided biases at the level of trial design by developing a protocol according to SPIRIT 2013 guidelines [48], preregistering the study, reporting even minor deviations, and attending to common sources of bias in clinical and prospective studies [49], though the possibility of unexplored confounding bias always must be considered [50]. Finally, 110 students in the control group indicated that they had seen a play or presentation by ACT Out! Ensemble before, though the degree of confounding influence, or the topic of prior performances, is unknown.

One such potential source of variance may have been the decision to randomize at the level of the classroom rather than at the level of the school. As has long been established, rigorous studies and evaluation of school-based programs is very methodologically difficult [51]. Prior research has indicated that randomizing at the school level can be problematic because schools often implement various SEL, SEC, and bullying programs, and so confounding variance can be introduced [23]. We were especially concerned about this when developing the protocol because SEL is written into expectations for Indiana schools, but not prescriptively (eg, schools can address it in different ways) [2]. By randomizing classrooms, we attempted to evade this problem by ensuring that a school’s other activities outside of this study were relatively equally represented among intervention and control clusters. However, it is important to consider that bullying does not occur within pre-existing clusters; that is, one is not limited to bullying or being bullied by students in the same homeroom period or English class. Thus, it is possible that an intervention affecting a bullying-related behavior, delivered to a random half of clusters within a grade and school, would have an effect on all clusters.

If this were the core mechanism explaining our data, then both bullying perpetration and victimization would theoretically be reduced; however, only victimization was reported to have been lowered. Thus, after careful consideration of the findings, including lack of SEC effect, high student receptivity and intervention fidelity, and significant time effects for traditional bullying victimization only, we hypothesize that the ACT Out! intervention may engender a heightened, defensive bystander response in participants. Bystander intervention in bullying has been associated with self-efficacy (belief that an intervention can be successful) [52], perceived knowledge about how one might successfully intervene [53], and the degree to which a bullying event is interpreted as serious [54]. Many of the scenarios for each grade level emphasized the roles of individuals other than the bully and the victim within the vignette and illustrated ways that others could intervene in a situation. They also identified potentially serious consequences that could emerge from bullying, including self-harm, so we infer surface-level plausibility of this explanation.

It is important to emphasize that while the data from this study were consistent with this explanation, the study itself does not provide causal evidence that the intervention engenders an increased likelihood for bystanders to intervene in bullying, nor was this an initial study hypothesis. Rather, we only know definitively that reported victimization decreased over time, in aggregate, among all participants. Additional research will be needed to determine the mechanism(s) by which this occurred. However, it is important to interpret and acknowledge all study findings to promote transparent research literature, and we have attempted to do so here.

This study was truncated unexpectedly by the COVID-19 pandemic, which had the effect of preventing planned 3-month outcome data collection and moderately affecting the number of clusters available for analysis for short-term outcomes (loss of 11 clusters). There were also several unplanned deviations from the study protocol, each of which has been documented in Multimedia Appendix 1. Participating schools were from both urban and rural counties in Indiana, and student participants were generally more diverse than the population of Indiana as a whole. However, some caution should be used when generalizing these findings outside of the participating schools, especially since participating schools were those that volunteered to participate in a randomized trial. Further, given sample proportions, the results can be generalized more readily to middle and high school students than elementary school students. The study also had several notable strengths, including prespecification of all analyses; use of multiple objective, external consultants to the research team; thorough documentation of all protocol deviations; use of validated measures; and provision of all student documents and data in an open-source format.

This study found no superiority for a 1-hour ACT Out! intervention compared to treatment as usual for SEC or offline bullying, but some evidence of a small effect for cyberbullying. As was already indicated, SEL and bullying interventions in schools tend to be lengthy and involved, and bullying interventions in particular may struggle to demonstrate effectiveness in randomized trials [24]. Since ACT Out! is much shorter and highly scalable, we interpreted the findings in this study, though few and small in magnitude, with interest.

We suggest several next steps for research in this area. First, a rigorous follow-up study with a new sample would be valuable, which addresses issues related to the interpretability of bullying victimization data, including measures of possible bystander effects and randomization at the school level while, if feasible, controlling for ongoing and recent bullying prevention programs. In doing so, scenario emphasis might also be reasonably shifted toward cyberbullying and away from physical bullying, for which potential iatrogenic effects in perpetration were computed in one of the models, though overall physical victimization declined. Second, additional data might also be collected on the sustainability of the effects beyond 2 weeks as well as on whether there is a dose-response relationship (eg, “Would 2 performances within a semester more strongly reduce victimization?”). This could also be extended by collecting measures related more broadly to student mental health, academic performance, and perceived school climate. Finally, on the practical side, given the high intervention fidelity, high student receptiveness, and preliminary evidence related to cyberbullying victimization, it would not be unreasonable for CMP to offer a performance of scenarios focused on cyberbullying prevention to supplement, rather than replace, extant bullying prevention programming. In practice (eg, outside of a trial), this intervention has comparatively low fiscal cost, only 1 hour of time is utilized, it requires no teacher time or preparation, and it may have some benefits.

To facilitate replication and transparent research processes, we have included supplemental files that may be valuable to researchers. These include a table of intracluster correlations (Multimedia Appendix 6), as well as the analysis syntax and datasets for per-protocol and intention-to-treat analyses (Multimedia Appendices 7-9, respectively).