استراتژی های مداخله کودکان در موقعیت های قربانی شدن توسط زورگویی: شناخت اجتماعی بیگانگان در مقابل مدافعان

Abstract This study examined the social cognitions of outsiders and defenders about intervening in situations of victimization by bullying. Do outsiders and defenders behave differently in victimization situations because of differences in competence beliefs, or because of a selectivity effect in intervening? These issues were examined in a sample of 102 outsiders and 107 defenders who were classified into these bullying roles through a peer-nomination procedure out of a total sample of 761 10- to 14-year-old Dutch children. These children were presented with imaginary victimization events. They answered questions about their cognitions and self-efficacy beliefs about intervening in victimization situations and about handling such situations. Outsiders, compared to defenders, claimed to intervene indirectly in victimization situations rather than directly. Defenders, compared to outsiders, claimed to intervene directly in victimization situations rather than indirectly. Both outsiders and defenders claimed to be more likely to intervene when a friend was being victimized than when a neutral classmate was being victimized. Outsiders and defenders did not differ in their self-efficacy for indirect intervention, but only defenders claimed a high self-efficacy for direct intervention. Both outsiders and defenders claimed to benefit from direct help when they themselves are victimized, but only outsiders also reported to need indirect help. The results suggest that outsiders and defenders behave differently in victimization situations because of differences in competence beliefs rather than because of a selectivity effect. More generally, the results suggest that not only defenders but also outsiders have the intention to help children who are being bullied. However, outsiders' anti-bullying attempts are likely to be indirect and less firm than those of defenders.

1. Introduction Victimization by bullying is generally defined as the repeated exposure to aggressive actions by one or more individuals over time (Greene, 2006, Olweus, 1993 and Salmivalli and Peets, 2009). During those actions some children (one or more bullies) intentionally attempt to harm another individual (the victim). Moreover, an imbalance in perceived or actual power exists between them. Bullying can be directly or indirectly aimed at a victim and can take the form of physical acts (e.g., hitting), possession-directed acts (e.g., taking belongings), verbal acts (e.g., name-calling) and social/relational acts (e.g., ostracizing; Espelage and Swearer, 2003, O'Connell et al., 1999 and Rigby, 2008). Bullying can be seen as a coercive strategy aimed at reaching and maintaining high ranks in the social hierarchy of the group (Olthof et al., 2011 and Reijntjes et al., 2013). Self-reports have been used to differentiate children as bullies, victims, bully–victims, and a group generally known as “bystanders” (Olweus, 1993 and Olweus, 2010). However, observational studies have demonstrated that the term bystander covers a wide range of different behaviors (Atlas and Pepler, 1998, Hawkins et al., 2001 and O'Connell et al., 1999). Bystanders include not only children who are uninvolved in the bullying process but also children who either side with the bully (assistants or reinforcers) or with the victim (defenders). Finally, there is a group of children who tend to shy away from the bullying; they turn their backs on the bullying. These children are referred to as the outsiders. Peer-reports such as those designed by Salmivalli, Lagerspetz, Björkqvist, Österman, and Kaukiainen (1996) enable the distinction between these participant roles. Meta-analytic studies suggest that anti-bullying interventions are more effective when they: (a) treat bullying as a group process (Polanin, Espelage, & Pigott, 2012); (b) focus on promoting prosocial bystander behavior (Polanin et al., 2012); (c) have a more intensive program (Ttofi & Farrington, 2011); or (d) target bullying from multiple points of view (e.g., targeting teachers and parents; Ttofi & Farrington, 2011). An example of a school-based anti-bullying intervention that incorporates these aspects in its program is KiVa (Salmivalli, Kärnä, & Poskiparta, 2010). One of KiVa's main aims is reducing bullying by influencing the different bystander-roles. Evaluation of KiVa has proven that it is effective in reducing bullying and victimization behaviors within classrooms (Kärnä et al., 2011). Although KiVa effectively reduces bullying and victimization, it is more effective in influencing teacher behavior than in promoting prosocial behavior in bystanders. Specifically with regards to pro-victim attitudes, including helping and intervening on behalf of the victim, the effect sizes of KiVa were small (Cohen's d's ≤ .08; Kärnä et al., 2011). Therefore, more knowledge of the different bystander roles is still needed to achieve further gains in the effectiveness of whole group anti-bullying interventions. The present study therefore aimed at improving the knowledge of children's social cognitions about intervening in victimization situations. The focus was on the bystander groups of outsiders (i.e., those who shy away from the bullying) and defenders (i.e., those who actively intervene on behalf of the victim). There are three reasons for focusing on outsiders and defenders. Firstly, observational studies indicate that peers are almost always present (more than 85% of the time) when bullying takes place (Atlas and Pepler, 1998, Hawkins et al., 2001 and O'Connell et al., 1999). Only in about a quarter of all victimization situations, peers actually discourage the bullying or act to alleviate the victim suffering caused by it. When peers decide to intervene on behalf of the victim, they do so effectively more than two-thirds of the time (Hawkins et al., 2001). The second reason for focusing on outsiders and defenders is that both have been implied to have an anti-bullying attitude. Previous findings have indicated that both outsiders and defenders are attitudinally against bullying and dislike children who bully others (Olthof and Goossens, 2008 and Salmivalli and Voeten, 2004). Defenders also behave in a way that matches their anti-bullying attitude; outsiders on the other hand do not behave according to their anti-bullying attitude—or at least not consistently enough to be qualified by their peers as defenders (Goossens et al., 2006, Salmivalli et al., 1996 and Sutton and Smith, 1999). The third reason for focusing on outsiders and defenders is their number. The group of outsiders has been found to be as large as one third of the classroom (Olthof et al., 2011 and Salmivalli et al., 1998). Another one-fifth of the classroom consists of defenders. Together they can make up more than 50% of the children in the classroom. Therefore, the sheer number of children, combined with their anti-bullying attitude makes them an attractive focus for interventions promoting defending behavior. A genuine reduction in bullying might only be achieved by urging children with an anti-bullying attitude to actively help victims and to stand up for their views (Gini, Pozzoli, Borghi and Franzoni, 2008, Orpinas and Horne, 2010, Polanin et al., 2012, Salmivalli, 1999 and Twemlow et al., 2010). Previous research has suggested that outsiders and defenders are quite similar to each other. Both are low in reactive and proactive aggression and are equipped with the ability to avoid being victimized themselves (Camodeca & Goossens, 2005). Next to that, both outsiders and defenders are relatively well liked by their peers (Goossens et al., 2006 and Salmivalli et al., 1996), which is an important motive for hanging out together and for endorsing similar prosocial behaviors (Duffy and Nesdale, 2009 and Witvliet et al., 2010). Moreover, outsiders and defenders have been found to be similar in empathy (Gini, Albiero, Benelli, & Altoè, 2008), an important prerequisite for prosocial behavior (Penner, Dovidio, Piliavin, & Schroeder, 2005). Research has indicated that children who are prosocial, are more likely to be friends with, and to be friendly to, victims. Conversely, having friends can be important for victims too, as friends protect each other when they are being victimized (Hodges et al., 1999 and Oh and Hazler, 2009).

3. Results 3.1. Preliminary analyses In Table 2 the results of the unconditional Intervention Preparedness model, the unconditional Intervention Efficacy model, and the unconditional Intervention Wish model are represented. As can be seen from Table 2, the total variance in the dependent variables was partitioned into components for the nested levels of child, classroom, and school. Partitioning the total variance at all levels, by calculating the intraclass correlations (ICC), indicated that the between-level variance in the outcomes was largely attributable to variability at the child level (Intervention Preparedness = 44.4%; Intervention Efficacy = 18.0%; Intervention Wish = 41.4%). Only small amounts of the variance in the outcomes could be attributed to the classroom or school level (Intervention Preparedness = 0.2% and 0.1%; Intervention Efficacy = 0.7% and 8.9%; Intervention Wish = 0% and 1.3% for the classroom level and school level, respectively). Table 2. Unconditional multilevel models for Intervention Preparedness, Intervention Efficacy, and Intervention Wish, of Outsiders versus Defenders. Intervention Preparedness Intervention Efficacy Intervention Wish F (df) Estimate F (df) Estimate F (df) Estimate Fixed effects parameters Intercept 5635.55⁎ (1, 13.63) 2.87⁎ 2697.31⁎ (1, 15.04) 3.00⁎ 3605.56⁎ (1, 17.20) 2.90⁎ Random effects parameters Child level variance 0.22⁎ 0.07⁎ 0.22⁎ Classroom level variance 0.00 0.00 0.00 School level variance 0.00 0.03 0.01 Deviance (− 2 log likelihood) 1590.49 751.98 882.66 Note. ⁎ p < .05. Table options In addition to the ICC, design effects were calculated for the nested levels as another indication of the need to take data clustering into account in model estimation. A design effect larger than 2 indicates that multilevel modeling is needed (Peugh, 2010). In the current study, the design effects for the classroom and school levels (Intervention Preparedness = 1.01 and 1.01; Intervention Efficacy = 1.03 and 2.00; Intervention Wish = 1.00 and 1.15) did not exceed this criterion. To acknowledge that the data had a nested structure, the classroom and school levels were specified in the analyses. The classroom and school level variances were small and nonsignificant (see also Table 2), and therefore, predictor effects were not modeled to vary across classrooms or schools.1 To test for the potentially confounding influences of Age (in months) and Ethnicity (Dutch, non-Dutch) on the dependent variables, these variables were added into each unconditional model. Neither had a statistically significant association with the dependent variables (p's > .05). The conditional models for every research question initially also included Age and Ethnicity as control covariates for effects on the independent variables. As these two control predictors did not influence the results, they will not be reported. 3.2. Intervention Preparedness The examination of differences in the cognitions of Outsiders and Defenders regarding their Intervention Preparedness in victimization situations and the influence of friendship with the victim on these cognitions started with the unconditional Intervention Preparedness model. As can be seen from Table 2, the unconditional Intervention Preparedness model estimated had a deviance statistic (− 2 log likelihood) of 1590.49. As a second step the independent dichotomous variables of Role, Intervention Strategy, Victim-type, Gender, and all interaction terms were added into the unconditional Intervention Preparedness model as fixed effects. A total of 15 predictors were added into the model. This full prediction model had a deviance statistic (− 2 log likelihood) of 1482.78, indicating that model fit was significantly worse without these predictors in the model, χ2(15) = 107.71, p < .001. Finally, all nonsignificant interaction terms were removed from the model to obtain the best fitting Intervention Preparedness model. In this final step, eight predictors were removed from the model, i.e., Role ∗ Gender, Victim-type ∗ Role, Victim-type ∗ Gender, all three-way interaction terms, and the four-way interaction term were removed from the model. The results of the final Intervention Preparedness model, including model fit statistics, can be found in Table 3. The unconditional model's fit was significantly worse than this model, χ2(7) = 102.26, p < .001. This set of predictors explained 6.51% of the total variance in Intervention Preparedness. Table 3. Final multilevel models for Intervention Preparedness, Intervention Efficacy, and Intervention Wish, of Outsiders versus Defenders. Intervention Preparedness Intervention Efficacy Intervention Wish F (df) Estimate F (df) Estimate F (df) Estimate Fixed effects parameters Intercept 5332.25⁎ (1, 14.23) 3.17⁎ 3155.34⁎ (1, 14.86) 3.16⁎ 3461.82⁎ (1, 20.21) 3.15⁎ Role 0.03 (1, 204.19) − 0.23⁎ 8.66⁎ (1, 209) 0.08 2.93 (1, 207.09) − 0.26⁎ Intervention Strategy 1.25 (1, 627) − 0.38⁎ 63.54⁎ (1, 209) − 0.61⁎ 1.39 (1, 209) − 0.15 Gender 1.38 (1, 200.06) − 0.30⁎ 3.66 (1, 208.67) − 0.01 8.51⁎ (1, 205.69) − 0.37 Victim-type 26.26⁎ (1, 627) − 0.06 Interaction terms Role ∗ Intervention Strategy 48.35⁎ (1, 627) 0.48⁎ 7.384⁎ (1, 209) 0.24⁎ 4.06⁎ (1, 209) 0.22⁎ Role ∗ Gender – – – – – – Intervention Strategy ∗ Gender 34.69⁎ (1, 627) 0.42⁎ 10.04⁎ (1, 209) 0.29⁎ 3.59 (1, 209) 0.21 Victim-type ∗ Intervention Strategy 10.32⁎ (1, 627) − 0.21⁎ Victim-type ∗ Role – – Victim-type ∗ Gender – – Random effects parameter Child level variance 0.23⁎ 0.11⁎ 0.21⁎ Deviance (− 2 log likelihood) 1488.23 665.57 866.63 χ2 difference test χ2(7) = 102.26⁎ χ2(5) = 86.41⁎ χ2(5) = 16.03⁎ Note. Gender was coded as 0 (Boys) and 1 (Girls). Role was coded as 0 (Defender) and 1 (Outsider). Intervention Strategy was coded as 0 (Direct) and 1 (Indirect). Victim-type was coded as 0 (Neutral) and 1 (Friend). The models were hierarchically built and represent the best fitting model. Blank cells in the fixed effects parameters section of the Table indicate parameters that were not present in the model. Dashed cells in the fixed effects parameters section of the Table indicate parameters that were not estimated in the final model due to nonsignificance in previous models. ⁎ p < .05. Table options There was a statistically significant interaction (see Table 3) between Role and Intervention Strategy (Role ∗ Intervention Strategy), F (1, 627) = 48.35; p < .001. This interaction indicated that Outsiders and Defenders were different in their cognitions regarding Intervention Preparedness. The descriptive statistics (means and standard errors of the mean) of the variables contributing to the interaction between Role and Intervention Strategy in Intervention Preparedness can be found in Table 4. Following-up on this interaction, Outsiders preferred Indirect over Direct Intervention, F (1, 306) = 40.19; p < .001, and were more prepared to intervene by Indirect means than Defenders, F (1, 209) = 5.02; p = .026. Defenders, on the other hand, preferred Direct over Indirect Intervention, F (1, 321) = 15.96; p < .001, and were more prepared to intervene by Direct means than outsiders, F (1, 207.63) = 8.01; p = .005. Table 4. Means (M) and standard errors of the mean (SE) for the interaction between Role (columns) and Intervention Strategy (rows) in Intervention Preparedness, Intervention Efficacy and Intervention Wish. Outsider Defender M SE M SE Intervention Preparedness Direct Intervention 2.72 0.06 2.97 0.07 Indirect Intervention 2.99 0.07 2.74 0.08 Intervention Efficacy Direct Intervention 2.68 0.06 3.01 0.07 Indirect Intervention 3.15 0.05 3.27 0.06 Intervention Wish Direct Intervention 2.93 0.07 2.92 0.08 Indirect Intervention 2.98 0.07 2.72 0.08 Table options Secondly, as can be seen in Table 3, a statistically significant main effect for the predictor Victim-type, F (1, 627) = 26.26; p < .001, was found as well as a statistically significant interaction with the predictor Intervention Strategy (Victim-type ∗ Intervention Strategy), F (1, 627) = 10.32; p = .001. These findings indicated that children's relationship to the victim influenced their cognitions regarding Intervention Preparedness. Children were less likely to intervene for a Neutral Victim (M = 2.77, SE = 0.04) than for a Friend as Victim (M = 2.94, SE = 0.04). This main effect was also found for both Direct Intervention, Mneutral = 2.71, SEneutral = 0.05; Mfriend = 2.98, SEfriend = 0.05; F (1, 209) = 94.26; p < .001, and Indirect Intervention, Mneutral = 2.83, SEneutral = 0.05; Mfriend = 2.91, SEfriend = 0.06; F (1, 209) = 7.64; p = .006. However, when children decided to intervene for a Neutral Victim, they were more likely to choose Indirect Intervention, F (1, 209) = 7.30; p = .007. When a Friend was victimized both strategies were chosen equally often (p > .05). Finally, a significant interaction between Gender and Intervention Strategy was found (Intervention Strategy ∗ Gender), F (1, 627) = 34.69; p < .001. This finding indicated that boys and girls were different in their cognitions regarding Intervention Preparedness. Boys preferred Direct Intervention (M = 2.90, SE = 0.07) over Indirect Intervention, M = 2.78, SE = 0.06; F (1, 234) = 11.72; p = .001. The reverse was found for girls, Mdirect = 2.71, SEdirect = 0.08; Mindirect = 3.02, SEindirect = 0.06; F (1, 384) = 29; p < .001. Moreover, girls were more likely to perform Indirect Interventions than boys, F (1, 209) = 8.48; p = .004. No significant gender differences were found for Direct Intervention (p > .05). Using the RML estimation method did not alter these results. 3.3. Intervention Efficacy The examination of differences in the cognitions of Outsiders and Defenders regarding their Intervention Efficacy in victimization situations, started with the unconditional Intervention Efficacy model. As can be seen from Table 2, the unconditional Intervention Efficacy model estimated had a deviance statistic (− 2 log likelihood) of 751.98. As a second step, the independent dichotomous variables of Role, Intervention Strategy, Gender, and all interaction terms were added into the unconditional Intervention Efficacy model as fixed effects. A total of 7 predictors were added into the model. This full prediction model had a deviance statistic (− 2 log likelihood) of 664.68, indicating that model fit was significantly worse without this set of predictors in the model, χ2(7) = 87.30, p < .001. Finally, all nonsignificant interaction terms were removed from the model to obtain the best fitting Intervention Efficacy model. In this final step, two predictors were removed from the model, i.e., the Role ∗ Gender interaction and the three-way interaction term were removed from the model. The results of the final Intervention Efficacy model, including model fit statistics, can be found in Table 3. The fit of the unconditional model was significantly worse than that of the final model, χ2(5) = 86.41, p < .001. The set of predictors explained 15.07% of the total variance in Intervention Efficacy. The results in Table 3 first of all show a main effect of the predictor Intervention Strategy, F (1, 209) = 63.54; p < .001. Children reported a significantly higher Indirect Intervention Efficacy (M = 3.21, SE = 0.04) than Direct Intervention Efficacy (M = 2.85, SE = 0.04). Secondly, Table 3 indicates that there was a statistically significant main effect of the predictor Role, F (1, 209) = 8.66; p = .004. Outsiders reported a lower Intervention Efficacy (M = 2.92, SE = 0.05) than Defenders (M = 3.14, SE = 0.05). However, a statistically significant interaction was also found between Role and Intervention Strategy, F (1, 209) = 7.384; p = .007. This interaction indicated that Outsiders and Defenders were different in their cognitions regarding Intervention Efficacy. The descriptive statistics (means and standard errors of the mean) of the variables contributing to the interaction between Role and Intervention Strategy in Intervention Efficacy can be found in Table 4. The lower Intervention Efficacy of Outsiders was only reflected in a lower Direct Intervention Efficacy, F (1, 208.35) = 15.11; p < .001. Outsiders were similar to Defenders in Indirect Intervention Efficacy (p > .05). Finally, Table 3 shows that there was a statistically significant interaction between Gender and Intervention Strategy, F (1, 209) = 10.04; p = .002. This finding indicated that boys and girls were different in their cognitions regarding Intervention Efficacy. Boys reported a higher Direct Intervention Efficacy (M = 3.01, SE = 0.07) than girls, M = 2.68, SE = 0.05; F (1, 208.71) = 10.50; p = .001. Boys and girls were found to be similar in Indirect Intervention Efficacy, Mboys = 3.23, SEboys = 0.06; Mgirls = 3.19, SEgirls = 0.05; p > .05. Still, both boys, F (1, 81) = 8.08; p = .006, and girls, F (1, 128) = 91.23; p < .001, reported a significantly higher Indirect Intervention Efficacy than Direct Intervention Efficacy. Using RML estimation method did not alter these results. 3.4. Intervention Wish The examination of differences in the cognitions of Outsiders and Defenders regarding their wish for help by others when they themselves were victimized started with the unconditional Intervention Wish model. As can be seen from Table 2, the unconditional Intervention Wish model estimated had a deviance statistic (− 2 log likelihood) of 882.66. As a second step, the independent dichotomous variables of Role, Intervention Strategy, Gender and all interaction terms were added into the unconditional Intervention Wish model as fixed effects. A total of 7 predictors were added into the model. This full prediction model had a deviance statistic (− 2 log likelihood) of 865.34, indicating that model fit was significantly worse without this set of predictors in the model, χ2(7) = 17.32, p = .02. Finally, all nonsignificant interaction terms were removed from the model to obtain the best fitting Intervention Wish model. In this final step, two predictors were removed from the model, i.e., Role ∗ Gender and the three-way interaction term. The results of the final Intervention Wish model, including model fit statistics, can be found in Table 3. Again, the fit of the unconditional model without predictors was significantly worse, χ2(5) = 16.03, p = .007. The final set of predictors explained 3.99% of the total variance in Intervention Wish. However, when using RML estimation, the two models' fit did not differ significantly and therefore these results are to be interpreted with caution. The interaction between Role and Intervention Strategy was statistically significant, F (1, 209) = 7.384; p = .045. This interaction indicated that Outsiders and Defenders differed in the way that they wished to be helped when they are being victimized. The descriptive statistics (means and standard errors of the mean) of the variables contributing to the interaction between Role and Intervention Strategy in Intervention Wish can be found in Table 4. Outsiders had a significantly higher wish for Indirect Help than Defenders, F (1, 209) = 5.87; p = .016, whereas Outsiders and Defenders were similar in their wish for Direct Help (p > .05). Moreover, Defenders had a lower wish for Indirect Help than for Direct Help, F (1, 107) = 5.43; p = .022, whereas Outsiders had a similar high wish for Direct and Indirect Help (p > .05). Finally, gender differences were also found. Specifically, a main effect was found for the predictor Gender, F (1, 205.69) = 8.51; p = .004. Boys (M = 2.76, SE = 0.07) had a lower Intervention Wish when being victimized than girls (M = 3.01, SE = 0.06). 4. Discussion The behavioral differences between outsiders and defenders in intervening on behalf of victims of bullying seem to be linked to their social cognitions. At the cognitive level, outsiders and defenders were found to differ because of competence rather than selectivity.