Empowering digital citizenship: An anti-cyberbullying intervention to increase children’s intentions to intervene on behalf of the victim

Cyberbullying rates among children are worrisome and the consequences of cyberbullying can be detrimental. Theory-based interventions to reduce cyberbullying are lacking. Therefore, this study examined whether an online anti-cyberbullying intervention based on the Theory of Planned Behavior could increase children’s intention to intervene in cyberbullying incidents on behalf of the victim. An experiment (N 1⁄4 298, 10–12 years old) was conducted to investigate whether the intervention could positively affect the behavioral intention to intervene on behalf of the victim via changes in children’s attitude, subjective norms, and perceived behavioral control. Results showed that children exposed to the anti-cyberbullying intervention had more intentions to intervene on behalf of the cyberbully victim compared to children who were exposed to a non-related intervention. However, no mediation effects were found for children’s attitude, subjective norms, and perceived behavioral control. The effectiveness of the intervention in the current study shows that it is important to develop theory-based intervention programs that also focus on the role of bystanders.

Bystanders have proven to play a crucial role in contesting bullying Kozubal, Szuster, & Barlínska, 2019). For example, bystanders can intervene by supporting victims, either directly (e.g., confronting the bully or comforting the victim) or indirectly (e.g., talking about the bullying incidents with adults; DeSmet et al., 2012;DeSmet, De Bourdeaudhuij, Walrave, & Vandebosch, 2019). In this way, the negative effects of cyberbullying experiences among victims may be reduced or bullying may even stop (DeSmet et al., 2019;Pepler, Craig, & O'Connell, 2010;Salmivalli, 2010). Previous research regarding traditional offline bullying has shown that interventions aimed to increase intervening behavior among bystanders were successful in reducing offline bullying (Hawkins, Pepler, & Craig, 2001;Salmivalli, Voeten, & Poskiparta, 2011). These promising results point to the relevance of investigating whether interventions are also effective in increasing positive intervening behavior (e.g., defending, comforting, reporting) in online bullying contexts. Eventually, this may lead to reduced cyberbullying prevalence rates (DeSmet et al., 2019Shakir et al., 2019). The current study examines whether an online anti-cyberbullying intervention can increase children's intentions to intervene on behalf of the victim in a cyberbullying context. So far, intervention studies have mainly focused on examining the effectiveness of anti-(cyber)bullying interventions among young adolescents (i.e., 15-19-year-olds;Cleemput et al., 2014;Espelage & Sung Hong, 2017). This study focuses on 10-to 12-year-olds, because research has shown that the prevalence rates and impact of cyberbullying is highly underestimated among this age group. Moreover, interventions to contest cyberbullying are lacking among this age group (De Castro et al., 2018;Snakenborg, Van Acker, & Gable, 2011;Tokunaga, 2010). Therefore, this study develops and evaluates an anti-cyberbullying intervention to increase intervening behavior on behalf of the victim. In addition, this study investigates whether changes in children's attitude, subjective norms, and perceived behavioral controlfollowing the Theory of Planned Behavior (Ajzen, 1991) explain this effect.

The theory of planned behavior in a cyberbullying context
Up to now, studies have focused primarily on the examination of determinants of cyberbullying behavior (e.g., frequency of ICT use, poor academic performance, and social support). In addition, previous studies have tried to develop anti-cyberbullying interventions targeting these determinants to reduce cyberbullying intentions and behavior (Calvete, Orue, Est� evez, Villard� on, & Padilla, 2010;Walrave & Heirman, 2011;Ybarra & Mitchell, 2004). Thus, the majority of studies have focused on empirically testing cyberbullying determinants while falling short in using a solid theoretical framework. As Snakenborg and colleagues stated (2011): "Most current cyberbullying programs are based on practical beliefs about prevention and logical approaches rather than on scientific theories" (p. 94). Furthermore, Tokunaga (2010) stated that "the indifference of cyberbullying researchers to make use of already established theories in new technology, mass media, and traditional bullying research is perplexing" (p. 285).
In the current study, we made use of the well-established Theory of Planned Behavior (TPB; Ajzen, 1991) to meet the calls of Snakenborg et al. (2011) and Tokunaga (2010).
The TPB states that people's intention to perform a certain behavior is the best predictor of their actual behavior. The behavioral intention, in turn, is predicted by three different concepts: (1) a person's attitude toward the behavior, (2) the subjective norm, which is a person's perception of what important others think of the behavior, and (3) the perceived behavioral control, which is the perceived ease or difficulty of performing the behavior.
Previous research examining cyberbullying behavior showed that the TPB is an effective and useful framework for studying cyberbullying behavior. For example, Heirman and Walrave (2012) studied whether the TPB could be used to predict adolescents' perpetration of cyberbullying. Overall, they found strong support for the theoretical utility of the TPB in cyberbullying research. The theory accounted for 44.8% of the variance in adolescent's behavioral intention to cyberbully. Specifically, the study results showed strong positive relationships between adolescents' attitudes, subjective norms and perceived behavioral control toward cyberbullying and their behavioral intention to perpetrate it. Pabian and Vandebosch (2014) corroborate the strength of using a TPB perspective in studying cyberbullying. In their survey study, they found that particularly a negative attitude and subjective norms toward cyberbullying resulted in lower intentions to cyberbully.
Although the TPB has shown to be a useful framework for studying cyberbullying intentions, these studies only examined the direct intention to cyberbully (Heirman & Walrave, 2012;Pabian & Vandebosch, 2014). It would be fruitful to study the intention to intervene in cyberbullying situations on behalf of the victims, because bystanders of cyberbullying could be crucial in reducing its prevalence rates Kozubal et al., 2019). Furthermore, when focusing on positive bystander behavior, an intervention targets a broader audience than cyberbullies or victims only, as anyone can witness cyberbullying (Pepler et al., 2010). Thus, children need to be provided with tools to help stop the cyberbullying behavior (DeSmet et al., , 2019(DeSmet et al., , 2012Pepler et al., 2010;Salmivalli, 2010). The current intervention provides these tools based on the three core concepts of the TPB. For example (see section 3.4 for detailed information), children are made aware a) of the importance to intervene (affecting their attitude) and b) that peers think it is important to intervene (subjective norms). In addition, c) they gain knowledge about how they can intervene (increasing their perceived behavioral control). Therewith, our intervention is expected to stimulate children's willingness to intervene: H1: Children who are exposed to the anti-cyberbullying intervention are more willing to intervene in cyberbullying situations compared to children who are exposed to a non-cyberbullying intervention.

Attitude toward intervening in cyberbullying situations
The TPB posits that a person's attitude toward certain behavior is based on one's behavioral beliefs (Ajzen, 1991). For example, research on cyberbullying showed that a more positive attitude toward cyberbullying was related to the belief that it provides a way to vent negative emotions, such as anger and revenge (K€ onig, Gollwitzer, & Steffgen, 2010). However, disapproval by peers, feelings of guilt, and knowing that it hurts the victim are associated with a negative attitude and a decline in cyberbullying intentions (Hinduja & Patchin, 2013;Menesini, Nocentini, & Camodeca, 2013). Importantly, previous research showed that behavioral beliefs associated with a negative attitude toward cyberbullying are related to a positive attitude toward intervening in cyberbullying situations to support the victim (McLaughlin, Arnold, & Boyd, 2005;Rigby & Johnson, 2006). Hence, we argue that in order to increase these positive attitudes toward intervening in cyberbullying contexts, children's knowledge and awareness concerning the negative consequences of cyberbullying among victims needs to be raised (Doane, Kelley, & Pearson, 2016;W€ olfer et al., 2013). Furthermore, it has been suggested that by inducing empathy toward victims of cyberbullying, children's attitudes toward cyberbullying should become more negative (Doane et al., 2016;W€ olfer et al., 2013). Consequently, by providing children with the knowledge and awareness of the negative consequences of cyberbullying and by raising empathy toward victims of cyberbullying during the intervention, we expect that children will intervene on behalf of the victim: H2a: Children who are exposed to the anti-cyberbullying intervention develop a more positive attitude toward intervening in cyberbullying situations compared to children who are exposed to a noncyberbullying intervention.
Furthermore, based on principles of the TPB (see Fig. 1, upper path), we also expect that: H2b: The effect of the intervention on the intention to intervene in cyberbullying situations is mediated by a positive change in attitude toward intervening in cyberbullying situations.

Subjective norms toward intervening in cyberbullying situations
The TPB posits that when someone thinks that important others think one should perform the behavior, it increases an individual's motivation to comply with the pressure exerted by these others (Ajzen, 1991). Research has shown that in the context of traditional bullying, perceptions about how significant others expected that one should behave predicted actual intervening behavior in a bullying context (McLaughlin et al., 2005;Rigby & Johnson, 2006). Studies on children and adolescents (i.e., 10-14 years old) showed that especially the expectations of friends on how to behave in offline bullying situations contributed significantly to children's actual behavior during bullying episodes (Mash & Wolfe, 2007;Rigby & Johnson, 2006). Therefore, it has been emphasized that children need to be provided with the knowledge that their peers believe (cyber)bullying is unacceptable and that it is important to intervene in cyberbullying situations (Doane et al., 2016). Furthermore, children should be made aware of their social responsibility to not intentionally harm others to create positive subjective norms within their peer groups (W€ olfer, 2013). By providing children with knowledge about the negative norms their peers hold toward cyberbullying and making them aware of their social responsibility during the intervention, we expect that: H3a: Children who are exposed to the anti-cyberbullying intervention develop more positive subjective norms toward intervening behavior in cyberbullying situations compared to children who are exposed to a non-cyberbullying intervention.
Furthermore, based on principles of the TPB (see Fig. 1, middle path), we also expect that: H3b: The effect of the intervention on the intention to intervene in cyberbullying situations is mediated by a positive change in subjective norms to intervene in cyberbullying situations.

Perceived behavioral control in cyberbullying situations
The TPB also states that perceived behavioral control plays a role in someone's intention to perform certain behavior (Ajzen, 1991). Previous studies have found that children often fail to intervene on behalf of a victim because of their lack of perceived ability to cope with the situation, and a lack of self-efficacy to intervene (Gini, Albiero, Benelli, & Altoe, 2008;Pepler, Craig, Ziegler, & Charach, 1994). Research shows that children are more likely to intervene in bullying situations when they know which actions they can take to stop the bullying and when they believe they have the resources to do so (Cramer, McMaster, Bartell, & Dragna, 1988). Therefore, researchers have argued that it is necessary to provide children with strategies on how to effectively intervene in online bullying contexts (Doane et al., 2016;W€ olfer et al., 2013). By providing children with the knowledge on how to effectively intervene in cyberbullying situations during the intervention we expect that: H4a: Children who are exposed to the anti-cyberbullying intervention will develop a higher perceived behavioral control to intervene in cyberbullying situations compared to children who are exposed to a noncyberbullying intervention.
Furthermore, based on principles of the TPB (see Fig. 1, lower path), we also expect that: H4b: The effect of the intervention on the intention to intervene in cyberbullying situations is mediated by a positive change in perceived behavioral control to intervene in cyberbullying situations.

Design
A repeated measures mixed-design experiment was conducted to investigate whether an online anti-cyberbullying intervention was effective in promoting intervening behavior on behalf of the victim among children between 10 and 12 years old. The between-factor in the experiment was condition, with the anti-cyberbullying intervention as experimental condition and a fake news intervention serving as the control condition. In another study, this fake news intervention served as the experimental condition while the anti-cyberbullying condition functioned as the control condition. The within-factor of the experiment was time with a pre-and post-exposure measurement. Children answered a number of questions using an online questionnaire before (pre-exposure) and three weeks after (post-exposure) the online intervention. The post-exposure measurement was scheduled three weeks later, to avoid that children would easily remember their answers to the pre-measurement questions. In addition, to check the manipulation of the stimulus materials (e.g., the online anti-cyberbullying intervention), children's knowledge, awareness, and empathy toward cyberbully victims were assessed immediately after being exposed to the intervention. The study received approval of the Ethics Committee of the Faculty of Social Sciences at Radboud University. The study is preregistered at the Open Science Framework (see osf.io/hcbw8).

Sample
Recruitment of participants involved a two-step procedure. First, schools were approached with the request to contribute to the research. In total, five schools across The Netherlands agreed to participate. After approval from the director of the school, information letters were spread among all parents/caretakers of the children in grades 5 and 6. Two information letters were created, one written for the anti-cyberbullying condition and one for the fake news condition. This letter informed parents about the goal and procedure of the study and the specific condition their children would be assigned to. It was emphasized that all information would be treated confidentially. Parents were asked to give active consent for participation of their children. Of all the parents, 52 (13.8%) did not gave permission. Prior to the start of the experiment, children were also asked whether they wanted to participate or not. Eight children did not want to participate. Furthermore, nineteen children were excluded because they were not present at either the pre-or the post-exposure measurement. The final sample consisted of 298 children (M ¼ 10.92, SD ¼ 0.70, range 10-12 years old), of whom 161 (54.0%) were girls.
A total number of sixteen classes participated in the experiment. Randomization took place at the school level: all participating classes within one school were assigned to either the anti-cyberbullying intervention (n ¼ 169; 48.5% girls, M age ¼ 10.93, SD age ¼ 0.70) or the fake news intervention (n ¼ 129; 61.2% girls, M age ¼ 10.90, SD age ¼ 0.71). This procedure assured that children from different classes within the same school were not able to exchange information about the different interventions with each other, which could have resulted in biased results. Power analysis using G*Power (Faul, Erdfelder, Lang, & Buchner, 2007) showed that a total of 286 children was sufficient to detect small effects (α ¼ 0.05, effect size ¼ 0.37, 90% power).

Procedure
The experiment took place in March and April 2019 during school hours. The experiment started with a short introduction to the study. After that, children received the assent form on which they had to sign whether or not they wanted to participate. After signing the assent form, children were directed to a website address they could find attached to the assent form. On this website, children started with the first questionnaire (pre-exposure measurement), which included general questions measuring their attitudes, subjective norms, perceived behavioral control and intentions toward intervening behavior in cyberbullying situations. We also asked some questions about their news consumption, their knowledge and awareness of fake news, and their self-efficacy to recognize fake news because this enabled us to also test the fake news intervention in another, separate study.
After completing the first questionnaire, children were automatically directed to either the anti-cyberbullying intervention or the fake news intervention (depending on the condition their school was assigned to).
Both the anti-cyberbullying intervention and the fake news intervention were online e-learning modules in which the children had to read information about either cyberbullying or fake news, complete tasks, answer questions, and view some videos. The duration of both interventions was approximately 35 min. After exposure to the intervention, they were automatically directed to the second questionnaire. This questionnaire captured questions about their knowledge, awareness, and empathy toward cyberbullying victims (manipulation check measures) and questions assessing their evaluation of the online intervention.
Three weeks after the intervention, children filled out the online post-exposure questionnaire during school hours. This questionnaire contained the same questions as the pre-exposure questionnaire. Additionally, we assessed children's social desirability biases to be able to control for this score in the main analyses, when necessary. This last questionnaire took approximately 15 min. After the last question, the researcher thanked the children for their participation. As a token of appreciation all participating children received a certificate. The debriefing procedure and presentation of the results took place after the data collection was completed via an information letter to both the school and the parents of the participating children.

Materials
Within the experimental condition, children were exposed to an online anti-cyberbullying intervention that aimed to increase children's intervening behavior in cyberbullying situations (see Fig. 2). Within the control condition, children were exposed to an online fake news intervention that was unrelated to cyberbullying. The fake news intervention was created in such a way that it was as similar as possible to the anticyberbullying intervention in terms of colors, navigation, number of questions, duration of the intervention, number of pages and number and duration of videos. Both online interventions were created using Gomo Learning software. Gomo Learning is a cloud-based responsive eLearning authoring tool that allows the user to create digital learning content.
The anti-cyberbullying intervention was created based on the Theory of Planned Behavior (Ajzen, 1991). Based on this theoretical model, we aimed to increase children's positive attitude toward intervening, positive subjective norms toward intervening, and their perceived behavioral control to intervene in order to increase their intentions to intervene in cyberbullying situations by addressing knowledge, awareness and empathy competencies. As previously mentioned, we specifically addressed knowledge, awareness, and empathy competencies since these constructs have proven to be highly important in changing people's attitude, subjective norms, and perceived behavioral control in bullying contexts (Espelage & Sung Hong, 2017;Williford et al., 2013). In order to positively change children's attitudes toward intervening behavior, children received knowledge about what cyberbullying actually is by providing the three-folded cyberbullying definition of Smith et al. (2008, p. 376): "An (a) intentional act carried out by (b) a group or individual, using (c) electronic forms of contact". Hence, in the intervention children were provided with knowledge about (1) the intentional character of cyberbullying, (2) the different groups that are involved in cyberbullying (the bully, the victim, and the bystanders), and (3) the online nature of cyberbullying. Furthermore, children's knowledge was targeted by providing them with a quick quiz containing three cyberbullying cases. For each of these cases they were asked whether the story described a cyberbullying situation or not. After children provided an answer, they received detailed feedback on why their answer was wrong or right. Moreover, in order to change children's attitudes, children were made aware of the negative consequences of cyberbullying. Also, their empathy toward victims of cyberbullying was targeted. We used two video clips for this (lasting 2.33 min in total). In these clips, victims talked about their distress and negative experiences during the time they were cyberbullied. Lastly, to change children's attitudes, we tried to increase empathy toward victims even further by specifically asking children to write down how they would feel when they would be cyberbullied.
In order to positively change children's subjective norms toward intervening behavior, children were asked to write down how many children they thought were bystanders of cyberbullying instances. After children wrote down a percentage, children were provided with knowledge about how many children were actual bystanders, namely 50% (Shultz, Heilan, & Hart, 2014), and how they themselves could be a part of this group. Furthermore, they were given knowledge about how children of their own age thought about intervening in cyberbullying incidents by mentioning the cool, brave, and important nature of intervening. Moreover, to increase children's subjective norms toward intervening behavior, we targeted children's empathy toward victims by exposing them to a video (lasting 1.22 min) displaying children of their own age talking about how they were cyberbullied. All these children conclude with the statement that it would have helped them a lot when their peers would have stand up for them to stop the cyberbullying.
In order to positively change children's perceived behavioral control toward intervening behavior, children were provided with knowledge about how they could intervene in cyberbullying situations. For example, during the intervention children were taught how they could try to comfort the victim, tell their parents or caretakers when something negative was going on online, or ask their teacher to help them when facing cyberbullying incidents among classmates. Lastly, children were exposed to a video (lasting 1.45 min) that provided them with knowledge on how they could intervene in cyberbullying, namely by posting positive comments online on the cyberbully victims' social media accounts.

Measures
The measures in this study either related to the manipulation check of the intervention (i.e., whether the intervention indeed targeted knowledge, awareness, and empathy) or to the main study (i.e., the dependent variable and potential mediators). Moreover, potential covariates were measured as well as socially desirable behavior.

Manipulation check measures
For the manipulation check, we investigated whether children who received the anti-cyberbullying intervention had indeed more knowledge and awareness of cyberbullying, while also having more empathy toward victims of cyberbullying compared to children who received the fake news intervention. To this end, children answered a few questions directly after exposure to either the online anti-cyberbullying intervention or the fake news intervention. Moreover, we measured their evaluation of the intervention to be able to provide some general insights into how they experienced the e-learning module.
Knowledge of cyberbullying. To assess children's knowledge toward cyberbullying, we adapted six items from Wahab and Yahaya's (2017) Cyber-bullying Knowledge and Awareness Instrument (CBKAi). Children answered on a six-point scale (1 ¼ Totally disagree and 6 ¼ Totally agree) regarding items such as: 'Creating a fake account on social media just for fun to target someone is cyberbullying' and 'Intentionally excluding a person from an online group (e.g., WhatsApp group) is cyberbullying'. A principal component analysis (PCA) of the six items was conducted. The analysis yielded one factor; all items loadings were >.6 on this factor (Cronbach's α¼ 0.824). We calculated a mean score of all the items assessing knowledge, representing the knowledge variable (M ¼ 4.83; SD ¼ 1.04).
Awareness of cyberbullying. To assess children's awareness of cyberbullying, we adapted five items from Wahab and Yahaya's (2017) CBKai. Children answered on a six-point scale (1 ¼ Totally disagree and 6 ¼ Totally agree) regarding the items: 'I think most victims of cyberbullying are anxious', 'I think most witnesses will intervene in cyberbullying situations', 'I think most victims of cyberbullying have the tendency to commit suicide', 'I think cyberbullying will stop if I do not intervene in cyberbullying', and 'I think most witnesses of cyberbullying report cyberbullying to someone else'. A PCA of the five items was conducted. The analysis yielded three factors; with all items loading > .6 on one of those factors. However, when creating mean scores for each of the factors, Cronbach's α was too low (α < 0.2) for all of the constructed variables. Therefore, we decided to include all items separately to perform the manipulation check.
Empathy toward victims. To assess children's empathy toward victims of cyberbullying, we used a seven-item empathy toward victim scale (P€ oyh€ onen, K€ arn€ a, & Salmivalli, 2008). Children answered on a six-point scale (1 ¼ Totally disagree and 6 ¼ Totally agree). The items were translated from an offline bullying context to a cyberbullying context. Items included for example: 'When the cyberbullied pupil feels sad, I want to comfort him/her' and 'I can see how the cyberbullied pupil is feeling bad'. A PCA of the seven items was conducted. The analysis yielded two factors; with all items loading > 0.6 on one of the two factors. The first factor represented affective empathy, consisting of items assessing children's emotions toward cyberbullied pupils (α ¼ 0.774). The second factor represented cognitive empathy, consisting of items assessing children's understanding of the emotions and needs of cyberbully victims (α ¼ 0.677). These two factors were also found in previous research on empathy and cyberbullying (Ang & Goh, 2010;P€ oyh€ onen et al., 2008). Therefore, we calculated mean scores for the items on each factor representing an affective empathy (M ¼ 3.79; SD ¼ 1.15) and cognitive empathy variable (M ¼ 4.85; SD ¼ 1.05).
Evaluation of the intervention. Besides the above-mentioned measures, children were asked to evaluate the intervention. They were asked to rate how interesting, important, fun, useless, and difficult they thought the intervention was on a visual analogue scale (0 ¼ totally disagree, 100 ¼ totally agree). Furthermore, they were also asked whether they had learned a lot from the intervention and whether they thought the intervention taught them something they did not know already.

Main study measurements
In order to test our hypotheses, several main measures were assessed, all at both pre-and post-exposure measurement.
Intention to intervene in cyberbullying situations. To assess children's intention to intervene in cyberbullying situations, we adapted a scale from an offline bullying context to a cyberbullying context (Stevens, van Oost, & de Bourdeaudhuij, 2000). We asked children whether, the next time they witness cyberbullying, they intent to (1) be an active bystander, (2) seek a teacher's help, (3) react against cyberbullies, (4) support the victim, and (5) spread the word that cyberbullying is stupid. Answers were measured using a visual analogue scale (0 ¼ Totally disagree to 100 ¼ Totally agree). A PCA of the five items was conducted for both the preand post-exposure measurement, yielding one factor at both time-points (see Table 1). After calculating mean scores and constructing the intention variables for both the pre-and post-exposure measurement, a difference score was created by subtracting the post-exposure measurement of intention from the pre-exposure measurement. This difference score was used in the main analyses (M ¼ À 0.79, SD ¼ 14.62).
Attitude toward intervening in cyberbullying situations. To assess children's attitude toward intervening in cyberbullying situations, we adapted items from the pro-bully and pro-victim scale of Stevens et al. (2000). The items were translated from an offline bullying context to a cyberbullying context. Children were asked to rate on a visual analogue scale (0 ¼ Totally disagree to 100 ¼ Totally agree) how much they agreed with the nine items. Examples of the items included: 'I understand children who cyberbully others' and 'Children who intervene in cyberbullying incidents are brave'. A PCA of the nine items was conducted for both the pre-exposure and post-exposure measurement (see Table 1). The analysis yielded three different factors at both measurement moments. Furthermore, three items loaded < 0.3 on one or more of the factors and three items double loaded on at least two factors at pre-exposure measurement. Moreover, the Cronbach's α of the remaining three items was too low at pre-exposure measurement, namely α < 0.164. At post-exposure measurement, two items loaded < 0.3 on one or more of the factors and two items double loaded on at least two factors.
Also, at post-exposure measurement the Cronbach's was low, namely α ¼ 0.481. Therefore, we concluded that we did not succeed in measuring children's attitude toward intervening in cyberbullying situations and where not able to calculate a mean score representing this attitude. We therefore decided not to analyze attitude as a mediating variable in the main analyses.
Subjective norms toward intervening in cyberbullying situations. To assess children's subjective norms toward intervening in cyberbullying situations, we adapted 6 items from the subjective norm scale of Sundstrom et al. (2018). We translated the items from a general bystander situation to a bystander situation in a cyberbullying context. We asked children whether their friends, teacher, parents, and others whose opinion matters to the them would support their decision to be an active bystander in a cyberbullying situation on a visual analogue scale (0 ¼ Totally disagree to 100 ¼ Totally agree). Furthermore, we asked whether their friends chose to be an active bystander in cyberbullying situations and whether they thought their friends would find them brave when they chose to be an active bystander. A PCA of the six items was conducted for both the preand post-exposure measurements, yielding one factor at both time-points (see Table 1). After calculating mean scores and constructing the subjective norms variables for both the pre-and post-exposure measurement, a difference score was created by subtracting the post-exposure measurement of subjective norms from the pre-exposure measurement of subjective norms. This difference score was used in the main analyses (M ¼ À 1.23, SD ¼ 16.02).
Perceived behavioral control to intervene in cyberbullying situations. To assess children's perceived behavioral control toward intervening in cyberbullying situations, we adapted items from the Defender scales (Salmivalli & Voeten, 2004). The items were translated from an offline bullying context to a cyberbullying context. We asked children on a visual analogue scale (0 ¼ Very difficult for me to 100 ¼ Very easy for me) how difficult or easy it would be for them to (1) try to make someone else stop the cyberbullying, (2) comfort the victim of cyberbullying, (3) encourage the victim to tell the teacher about the cyberbullying, and (4) choose to be an active bystander when someone is cyberbullied. A PCA of the four items was conducted for both the pre-and post-exposure measurements, yielding one factor at both time-points (see Table 1). After calculating mean scores and constructing the perceived behavioral control variables for both the pre-and post-exposure measurement, a difference score was created by subtracting the post-exposure measurement of perceived behavioral control from the pre-exposure measurement of perceived behavioral control. This difference score was used in the main analyses (M ¼ 1.17, SD ¼ 17.09).

Covariates
Children's sex, age, and grade were assessed as potential covariates. Moreover, we assessed children's media use and social desirability as potential covariates.
Media use. Children's media use was assessed as potential covariate. Children indicated how often they used (1)  Social desirability. Previous research has shown that cyberbullying research is highly vulnerable to social desirability biases . The potential factor of socially desirable behavior can lead to over-or under-reporting of cyberbullying and/or victimization of cyberbullying. Generally, children consider cyberbullying as socially undesirable behavior and standing up for a victim as socially desirable behavior (Akbulut & Eristi, 2011;Betts, 2016). These considerations lead studies to find a positive association between cyberbullying and social desirability (Doane, Kelley, Chiang, & Padilla, 2013). It is, therefore, important to control for social desirability.
We assessed children's social desirability biases using the adapted 7item Children's Social Desirability Short (CSD-S) scale from Miller et al. (2015). Items included: 'Do you always listen to your parents?' and 'Have you ever broken a rule?'. Children answered on a six-point scale (1 ¼ Totally disagree to 6 ¼ Totally agree). The Kaiser-Meyer-Olkin measure verified sampling adequacy (0.750), and the Bartlett's test of sphericity agreed that the correlations between the items were large enough to conduct a PCA: χ2 (10) ¼ 453.19; p < .001. Therefore, a PCA of the seven items was conducted. The analysis yielded one factor (α ¼ 0.773); all items loading were > 0.6 on this factor. Therefore, we calculated a mean score combining all the items in one variable measuring social desirability (M ¼ 2.98; SD ¼ 1.08).

Strategy of analysis
Before we started to analyze the hypotheses, we conducted some preliminary analyses. Then, a randomization check was performed using ANOVA to see whether the two conditions significantly differed from Note. The mean scores of the pre-and post-exposure measures did not significantly differ for intention to intervene (p ¼ .351, subjective norms toward intervening (p ¼ .185), and perceived behavioral control toward intervening (p ¼ .239). each other with respect to age, sex, and grade. Second, we investigated by means of a manipulation check whether children who received the anti-cyberbullying intervention had indeed more knowledge and awareness of cyberbullying while also having more empathy toward cyberbully victims compared to children who received the fake news intervention. We used t-tests to perform the manipulation check. Lastly, to look for potential covariates, correlations were performed between all study variables. All main analyses were performed using Hayes' PROCESS macro for mediation models in SPSS (Hayes, 2018). This macro provides an analysis of the mediation effects hypothesized, following the causal steps approach (Baron & Kenny, 1986). All the hypotheses were tested using Model 4. We performed a within-subject analysis using PROCESS to be able to see whether the constructs of the Theory of Planned Behavior positively changed over time. In order to do this, we entered the independent variable (condition), the two difference scores of the mediator variables (subjective norms and perceived behavioral control), and the difference score of the dependent variable (intention to intervene), and the control variables (see section 4.2.3 for more information) simultaneously in a model 4 PROCESS model. Bootstrap confidence intervals (5000 bootstrap samples, 95% bias-corrected confidence intervals) were produced to test the significance of the direct and mediated effects (Hayes, 2018). Bootstrap confidence intervals are preferred over p-value testing because bootstrapping respects the non-normality of the sampling distribution of the indirect effects.

Descriptive statistics
Descriptive statistics showed that 32.2% of the children made use of social media every day, whereas 27.2% of the children reported to never use social media. Furthermore, 44.0% of the children made use of WhatsApp every day, opposed to 16.8% of the children who never used WhatsApp. On average, children believed that 70.3% of all bystanders actually intervene in cyberbullying situations, which is higher than it actually is (50%; Shultz, Heilamn, & Hart, 2014). With respect to the evaluation of the intervention, descriptive statistics showed that children who received the anti-cyberbullying intervention thought the intervention was interesting (

Randomization check
A randomization check showed that the two conditions did not differ with regard to age (F (1, 296) ¼ 0.131, p ¼ .718) and grade (F (1, 296) ¼ 0.121, p ¼ .483). However, the two conditions did differ with regard to sex (F (1, 296) ¼ 4.81, p ¼ .029). Therefore, sex was added as a control variable in the analyses.
Children in the experimental condition showed more awareness of cyberbullying compared to children in the control condition. Children in the experimental condition were more aware of the suicide attempts among victims (M ¼ 3.65, SD ¼ 1.16) compared to children in the control condition (M ¼ 3.09, SD ¼ 1.33), t (296) ¼ À 3.86; p < .001. They were also more aware of victims anxiousness (M ¼ 5.51, SD ¼ 1.00) compared to children in the fake news intervention (M ¼ 5.06, SD ¼ 0.99), t (296) ¼ À 3.82; p < .001. Moreover, they showed more awareness with respect to the lack of bystanders intervening on behalf of cyberbully victims (M ¼ 3.02, SD ¼ 1.18) compared to children in the control condition (M ¼ 2.61, SD ¼ 1.18), t (296) ¼ À 2.98; p ¼ .003. Lastly, children receiving the anti-cyberbullying intervention were also more aware of the fact that most bystanders do not talk about cyberbullying with others (M ¼ 5.08, SD ¼ 1.30) compared to children in the control condition (M ¼ 4.55, SD ¼ 1.59), t (296) ¼ À 3.19; p ¼ .002. Overall, we can conclude based on these t-tests, that the manipulation between the anti-cyberbullying intervention and the fake news intervention was successful.

Covariates
With respect to potential covariates, exploratory analyses showed no significant correlations between age, sex, internet use, social media use, Whatsapp use and the dependent variables. However, social desirability was significantly correlated with subjective norms toward intervening (r ¼ .211, p < .001), perceived behavioral control toward intervening (r ¼ 0.111, p ¼ .025), and intention to intervene (r ¼ 0.203, p < .001). Therefore, the variable social desirability was included as a control variable in the main analyses. Besides, based on the randomization check mentioned earlier, we also added sex as a control variable to the main analyses.

Main analyses
The first hypothesis predicted that children who were exposed to the anti-cyberbullying intervention would become more willing to intervene in cyberbullying situations compared to children who were exposed to a non-cyberbullying intervention. This hypothesis was supported. The results (see Fig. 3) showed a direct effect of condition on intention to intervene on behalf of a cyberbully victim (B ¼ 3.45, boot SE ¼ 1.62, BC 95% CI [0.26, 6.63]).
Hypothesis 3a predicted that children who were exposed to the anticyberbullying intervention would develop more positive subjective norms toward intervening compared to children who were exposed to the fake news intervention. Results showed that this hypothesis was not supported (B ¼ .91, SE ¼ 1.92, BC 95% CI [-2.87, 4.68], see Fig. 3). Furthermore, we expected that the effect of the intervention on children's intention to intervene in cyberbullying situations was mediated by a positive change in subjective norms to intervene in cyberbullying situations (H3b). Results showed that the direct effect of the children's subjective norms on their intentions to intervene was significant (B ¼ 0.25, boot SE ¼ 0.05, BC 95% CI [0.15, 0.35]). However, there was no indirect effect of the intervention on children's intentions to intervene via subjective norms (effect ¼ 0.23, boot SE ¼ 0.49, CI [-0.77, 1.18], see Fig. 3). Therefore, hypothesis 3 b was not supported.
In hypothesis 4a, we predicted that children who were exposed to the anti-cyberbullying intervention would develop a higher perceived behavioral control toward intervening in cyberbullying situations compared to children who were exposed to the fake news intervention. Results showed that this hypothesis was not supported (B ¼ 2.12, boot SE ¼ 2.05, BC 95% CI [-1.91, 6.15], see Fig. 3). We also tested whether the effect of the intervention on intention to intervene in cyberbullying situations was mediated by a positive change in perceived behavioral control to intervene in cyberbullying situations (H4b). Results showed that the direct effect of children's perceived behavioral control on their intentions to intervene was significant (B ¼ 0.16, boot SE ¼ 0.05, BC 95% CI [0.07, 0.26]). However, the results showed that the indirect effect of the intervention on children's intentions to intervene via their perceived behavioral control was not significant (effect ¼ 0.34, boot SE ¼ 0.39, CI [-0.25, 1.28], see Fig. 3), implying that the hypothesis was not supported.

Post-hoc analyses
Results of the main analyses showedin contrast to what we expectedthat there were no mediation effects of children's subjective norms and perceived behavioral control on their intentions to intervene in a cyberbullying context. However, the results of the manipulation check showed that knowledge, awareness, and affective empathy were successfully targeted in the intervention. Therefore, we performed posthoc analyses to test whether knowledge, awareness, and affective empathy might be the underlying mechanisms explaining the direct effect of the intervention on the positive increase in children's intentions to intervene on behalf of a victim. The post-hoc analyses were performed following the same procedure as the main analyses, using Hayes' PRO-CESS analyses Model 4 (Hayes, 2018).
The results of the post-hoc analyses showed that the direct effect of the anti-cyberbullying intervention on children's intentions to intervene was not mediated by children's knowledge of cyberbullying (B ¼ 0.11, boot SE ¼ 0.23, BC 95% CI [-0.28, 0.66]), their awareness of cyberbullying (B ¼ 0.04, boot SE ¼ 0.14, BC 95% CI [-0.27, 0.36]) or their affective empathy toward cyberbully victims (B ¼ 0.03, boot SE ¼ 0.40, BC 95% CI [-0.79, 0.81]). Thus, we are not able to explain which factors contributed to the direct effect of the intervention on children's intentions to intervene.

Discussion
In the current study, we developed and evaluated an anticyberbullying intervention that aimed to increase children's intervening behavior on behalf of the victim. The anti-cyberbullying intervention was successful in stimulating children's intentions to intervene in cyberbullying incidents. Surprisingly, we could not explain this effect by the three concepts of the Theory of Planned Behavior nor via knowledge, awareness, or affective empathy. Nevertheless, the increase in intentions to intervene among children who were exposed to the cyberbullying e-learning module is important for the field of cyberbullying prevention. In particular, the findings demonstrate the promise of low-cost, brief (35 min) online cyberbullying intervention programs for children.
Our findings showed that children who perceived higher positive subjective norms had a higher intent to intervene in online bullying contexts. In addition, we found that higher perceived behavioral control was associated with higher intentions to intervene in cyberbullying situations. Although these findings are in line with the TPB and previous research (Heirman & Walrave, 2012;Pabian & Vandebosch, 2014), these concepts did not serve as mediators. For subjective norms, involving significant others in prevention programs to contest cyberbullying might be important. In the current intervention, children mainly received written information about how many children of their age intervene in cyberbullying incidents, and how they themselves can be part of this group. Previous research (Heirman & Walrave, 2012) suggests that it could be more effective to actually expose children to how their own peers think about intervening in order to positively change their subjective norms. For example, children can be exposed to a video during the intervention showing peers rejecting cyberbullying behavior and arguing in favor of intervening on behalf of a victim.
With regard to perceived behavior control, the finding that the current intervention was not successful in targeting cognitive empathy could have played a role in the absence of a mediating effect. Previous research showed that children need to be able to understand the feelings and needs of a cyber victim (e.g., experiencing cognitive empathy) in order to feel capable of helping this person (Barlinska, Szuster, & Winiewski, 2018). To this endfollowing the recommendations of Barlinska et al. (2018) -, future research could expose children to videos of cyberbully victims expressing their needs and how bystanders can fulfill these in order to strengthen children's perceived behavioral control.
A related issue regarding our intervention is that the nature of what constitutes cyberbullying can be ambiguous. Youth typically do not conceptualize cyberbullying as it is defined. For example, their definitions almost always omit the components of intentionality and repetition (Vaillancourt et al., 2008). It is because of this ambiguity that children could experience difficulty with interpreting an event as a call to action. To be more specific, it is possible that children did not perceive the relevant cues (e.g., hearing derogatory names), which may have hindered them from noticing events as cyberbullying episodes (Dinkes, Kemp, Baum, & Snyder, 2009;Loewestein & Small, 2007). It is recommended to incorporate more vivid elements in future intervention studies.
With regard to our theoretical framework, it may be useful to further investigate anti-cyberbullying interventions in light of another theoretical framework that comprises the concepts of the TPB. For example, Fig. 3. Standardized regression coefficients for the relationship between the anti-cyberbullying intervention and intention to intervene as mediated by positive subjective norms and higher perceived behavioral control using difference scores. *p < .05; **p < .001. the bystander intervention model of Latan� e and Darley (1970) describes five steps one must take in order to become an active bystander during an event. People have to (1) notice the event, (2) interpret the event as an emergency that requires help, (3) accept responsibility for intervening, (4) know how to intervene or provide help, and (5) implement intervention decisions. The three last steps of this model were incorporated in our intervention program, whereas the first and second step were not. Although we already found the intervention to be successful by focusing on steps three to five, the bystander intervention model suggests that step 1 and 2 are prerequisites for an individual to become an active bystander (Latan� e & Darley, 1970). Future research should investigate whether the intervention effect is even stronger when incorporating the first two steps. In addition, one should test which specific aspects (e.g., providing information, exposing children to videos, let children perform tasks) are particularly effective in a cyberbullying context to make explicit recommendations for future developments of interventions.
Another factor that may have contributed to not finding any mediation effects could be the three-week timespan between pre-and postexposure measurement. For example, children could have experienced cyberbullying situations in the meantime either directly (e.g., being victimized) or indirectly (e.g., witnessing someone being cyberbullied). By experiencing cyberbullying incidents, it could be that children were not confident anymore in their abilities to intervene (e.g., perceived behavioral control). Furthermore, it could be that children saw that peers did not intervene in online bullying incidents, which could have withheld the intervention from positively changing children's subjective norms.
Although the TPB has proven to be a useful framework for studying cyberbullying related behaviors (Heirman & Walrave, 2012;Pabian & Vandebosch, 2014), an important limitation needs to be mentioned. The theory has been criticized for excluding subconscious processes as important influences on behavior (Sheeran, Gollwitzer, & Bargh, 2013;Sniehotta, Presseau, & Araújo-Soares, 2014). Especially with regard to intervening in cyberbullying incidents, these processes might be relevant to investigate because several studies have shown that subconscious processes may come into play when finding oneself in a bystander position during bullying and cyberbullying incidents (Carter, M'Balla-Ndi, Van Luyn, & Goldie, 2016;Van den Bos, Müller, & Van Bussel, 2009). For example, the subconscious process of audience inhibition has proven to play an important role in bullying bystander behavior. When finding oneself in a bystander situation, audience inhibition posits that someone is subconsciously driven to not intervene in such a situation as a result of non-intervening others (Carter et al., 2016). Because subconscious processes can play an important role in explaining cyberbullying and intervening behavior, future research should include implicit behavioral measures to further explore these processes.
This study had some practical limitations. Although the intervention was effective as a whole in positively changing children's intentions to intervene, it may be that specific aspects of the intervention were especially effective in accomplishing this positive change. However, this change remained unknown because children's knowledge, awareness, and empathy as a result of the intervention were only measured after the intervention. Therefore, we were not able to give insight into the possible effect on the intervention itself nor the relationship between these changes and the mediators of the underlying theoretical model of this study. With respect to future research, it would be valuable to account for children's changes in knowledge, awareness, and empathy. These insights could provide explanations for the effect found in our study, namely a positive change in children's intentions to intervene on behalf of a cyberbully victim due to exposure to the anti-cyberbullying intervention.
Second, this study did not include past cyberbullying behavior and/ or cyberbullying victimization because of the potential negative psychological effects when confronting children with questions about cyberbullying and/or cyberbullying victimization experiences in their past. However, it could be that the effectiveness of the intervention differs for bully victims and bully perpetrators (De Castro et al., 2018). For example, previous research showed that victims of both traditional bullying and cyberbullying are more inclined to help another victim, because they sympathize more with this person (Fawzi & Goodwin, 2011). On the contrary, being a cyberbully has proven to be an important predictor of neglecting to intervene on behalf of a victim (Fawzi & Goodwin, 2011;Szuster, 2016). Including past cyberbullying behavior might be relevant to investigate the potential differential effects between victims and bullies with respect to the effectiveness of interventions.
Third, we were not successful in capturing the construct attitude toward intervening behavior. During the experiment, children asked for clarification on the attitude items in the questionnaire several times. It is plausible that we could not construct a reliable and valid scale for attitude because the children did not understand the questions. The items used in the questionnaire originated from an existing scale targeted at adolescents (Stevens et al., 2000). It could be that the items were clear to adolescents, but that our target group was too young to comprehend the items at hand. Future research should therefore try to develop an attitude scale measuring intention to intervene specifically targeted at children.
Another limitation is that all findings rely on children's self-reports which are vulnerable to social desirability biases. However, due to characteristics of cyberbullying such as its invisibility and anonymity, it is extremely difficult to measure cyberbullying related behaviors using objective measures (W€ olfer et al., 2013). We tried to address this limitation by assessing and controlling for children's potential social desirability biases. Also, in the current study we did not assess children's actual intervening behavior. However, based on the TPB and previous research that examined the relationships between the constructs of the TPB in a cyberbullying context, it is expected that the increase in children's intentions to intervene actually leads them, at least to a certain extent, to intervene more in cyberbullying situations (Heirman & Walrave, 2012;Pabian & Vandebosch, 2014). To address both the social desirability issues and including actual intervening behavior, future research could follow the recommendations of Dillon and Bushman (2015), by exposing children to online cyberbullying incidents in a mediated environment in which one confederate acts as a cyberbully and another as a victim. In this way, the actual intervening behavior of the child can be monitored during the experiment.
Although scholars agree on the need for effective interventions to increase intervening behavior and to reduce cyberbullying rates (Snakenborg et al., 2011;Tokunaga, 2010), some scholars raised questions about the possible side-effects of such interventions on especially (cyber)bully victims. For example, De Castro et al. (2018) showed that some widely used large scale intervention programs produce counter effective results, leading to more (cyber)bully conflicts instead of less, and possible negative affect in children receiving the intervention. Nevertheless, we argue that many interventions Stevens et al., 2000;Wolfer et al., 2013), including the one in the current study, have substantial positive effects with respect to contesting cyberbullying, even after a short, one-time exposure. Furthermore, children in the current study did not express negative affect regarding the intervention itself. They really enjoyed working on the e-learning and found it very interesting.
To conclude, this study adheres to the call of Snakenborg et al. (2011) andTokunaga (2010) to develop an anti-cyberbullying intervention based on solid theoryin the present study the Theory of Planned Behaviorto increase children's intervening behavior. This is, to our knowledge, the first comprehensive, theory-based cyberbullying intervention aimed at increasing positive bystander behavior in children. Previous bullying and cyberbullying interventions showed mixed effects with respect to their effectiveness to reduce (cyber)bullying (De Castro et al., 2018). The effectiveness of the intervention in the current study points to the relevance of developing theory-based intervention programs that focus on the important role of bystanders besides targeting bully's or victims (DeSmet et al., , 2012Gini et al., 2008). Given the high prevalence rates concerning cyberbullying among children and its negative consequences, it is of utmost importance to keep developing and empirically evaluating intervention programs to reduce cyberbullying. We believe the current online anti-cyberbullying intervention has the potential to guide future intervention approaches by providing teachers and schools with helpful educational tools to empower children to stand up against cyberbullying.

Declaration of competing interest
No declaration of interest.