Introduction

Since the landmark studies of Salmivalli and colleagues in the 1990s on participant roles in bullying episodes, bullying has been recognized as involving not only the bully and the victim but also various bystanders, such as assistants (e.g., following the bully), reinforcers (e.g., laughing at what is happening), defenders (e.g., being supportive of the victim), and outsiders (e.g., staying away from the bullying situation altogether) (Salmivalli et al. 1996). We followed the accepted definition of a bystander as a person who witnesses bullying but is not involved as a bully or victim (Cowie 2000; Smith et al. 1999). A bystander may continue to passively observe a bullying episode, start to encourage a bully, or actively defend a victim. Considering that most bullying occurs in an environment where peers are witnessing the incident (Hawkins et al. 2001), both the bully’s personal motivation and a bystander’s action or inaction could be influenced by the quest for high status. That is, bullying is not merely based on the power imbalance between bully and victim (Olweus 1993), but it is also a group process driven by the social needs of the involved peer groups. Whereas bullies have a social need to achieve high status, bystanders may be motivated to fit in with the peer group and avoid becoming a target of the bully themselves (Buhrmester 1990; Jarvinen and Nicholls 1996). In this article, we focused on defenders as one relatively neglected bystander role and what factors might determine their willingness to come to the aid of the victim.

Although most youth display attitudes against bullying and report a willingness to help victimized peers in hypothetical situations (Boulton et al. 2002; Rigby and Johnson 2006), actual defending behavior is rare. For example, one observational study reported that only about 10–17% of students actually stood up for victims of bullying (Pepler and Craig 1995). These results suggest that solely measuring students’ attitudes against bullying or in support of victims is not sufficient for understanding defending behavior. Because bullying is a peer group process involving various kinds of participant roles that are enacted within the school context (Salmivalli et al. 1996), school contextual factors such as group norms as well as interpersonal factors such as social status in the peer group need to be evaluated as possible determinants of defending behavior. To build on previous studies, the research presented here adopts a multilevel approach to examine how intrapersonal, interpersonal, and contextual factors are associated with defending behavior.

Empathy as an Intrapersonal Factor

Traditionally, empathy has been regarded as a cornerstone in motivating prosocial behavior (e.g., Hoffman 2000). Since Feshbach (1978) first proposed empathy as a multidimensional construct, a number of studies have consistently shown that empathy has two main components: cognitive empathy and affective empathy (e.g., Davis 1994). Cognitive empathy is defined as the capacity to take another person’s perspective, while affective empathy represents sharing others’ emotions and feeling concern or sympathy toward others (Feshbach 1978). When making a distinction between the cognitive and affective components, some studies found that affective empathy, rather than cognitive empathy, was associated with defending behavior (e.g., Barchia and Bussey 2011; Caravita et al. 2009; Peets et al. 2015; Pöyhönen et al. 2010). On the other hand, other studies showed that defending was positively associated with cognitive empathy or both types of empathy (e.g., Gini et al. 2008).

In terms of gender differences, girls report higher levels of empathy (Eisenberg et al. 1998; Olweus and Endresen 1998) and tend to display more prosocial behavior than boys (Rose and Rudolph 2006). In studies examining associations between empathy and willingness to defend the victim, empathy was a critical factor in predicting defending behavior for girls, while there was no direct association between empathy and defending behavior for boys (Barchia and Bussey 2011). In one study, adolescent boys showed a positive association between affective empathy and defending behavior, but only if they were well-liked in their peer groups (Caravita et al. 2009).

Perceived High Social Status as an Interpersonal Factor

Because bullies are often perceived as popular in the peer group especially during adolescence (Juvonen et al. 2003), defending victims of bullying might be a special type of prosocial behavior that carries its own risks. Even though potential defenders are empathetic, if they are not also of high enough status to confront bullies, defending behavior may backfire and increase the defender’s risk of being a future target of bullying. This fear of being victimized is a critical factor contributing to passive bystander behavior rather than defending during bullying incidents (Thornberg et al. 2012). Potential defenders who are highly regarded by their peers are probably in the best position to manage that fear.

Although social status appears to be a critical factor in motivating students to come to the aid of their victimized peers, only a few studies have considered this factor in defending behavior and the findings have been inconsistent. For example, Pöyhönen et al. (2010) documented a positive association of both likability and popularity with defending behaviors, whereas Caravita et al. (2009) found the same only for likability. It is evident that more research is needed on the role of social status as a predictor of defending behavior.

School-level Prosocial Norms as a Contextual Factor

Not all individuals who have a high level of empathy and high social status in the peer group are inclined to defend victimized peers (Peets et al. 2015). An important but understudied factor might be the normativeness of defending behavior in one’s classroom or school. For example, do other students share the belief that classmates ought to come to the aid of the victim? Since peers motivate individuals to engage in prosocial behavior (Bandura 1986), peers’ behaviors or attitudes in classroom or school contexts create peer norms that shape individuals’ prosocial behaviors. These peer norms are powerful regulators of behavior in peer contexts, especially in adolescence (Brechwald and Prinstein 2011; Prinstein and Wang 2005; Veenstra et al. 2013).

Because bullying research on group norms has focused on aggressive behavior norms (e.g, Dijkstra et al. 2008; Peets et al. 2015; Salmivalli and Voeten 2004; Sentse et al. 2007), we know less about whether norms can influence defending behavior as well (Pozzoli et al. 2012). We located only one study that investigated both peers’ prosocial behaviors and attitudes as contextual factors (Pozzoli et al. 2012). That study found that students were more likely to defend victimized peers in classrooms where the level of prosocial behaviors or attitudes was high.

The Need for a Multilevel Approach

Researchers have increasingly recognized the importance of promoting defending behavior to decrease bullying (Salmivalli 2014), yet the field knows relatively little about the factors that motivate someone to come to the aid of a victim. We adopted an ecological systems approach (Bronfenbrenner 1979) to propose that the interaction between individual and contextual factors can shed light on factors that encourage or discourage specific behaviors in bullying situations (e.g., Espelage 2014). We took as a starting point that individual level factors such as empathetic concern would be important motivators of defending behavior. However, because bullying is a peer group process involving various participant roles in classrooms or schools (Salmivalli et al. 1996), we proposed that defenders might need both high social status in the peer group to risk confrontation with bullies, who often enjoy their own popularity (notoriety) during adolescence, and peer norms in classroom or school contexts that support defending behavior. Few studies on defender behavior to date have analyzed these individual, social, and contextual factors simultaneously in a multilevel framework that allows for tests of person-context interactions. In addition, most bystander studies have focused on white students in European countries (e.g., Finland, Italy, and the Netherlands), so little is known about which factors are associated with defending behavior in other parts of the world, including the defending behaviors of youth in Asian countries.

To expand the cross-cultural breadth of studies on predictors of defending behavior in early adolescence, we focused on South Korea as a research context for two reasons. First, South Korea has been characterized as a collectivist culture in which students tend to conform to peer group over individual attitudes and behaviors (Farver et al. 2000; Hong et al. 2014). Korean bullying, called Wang-ta, has strong conformity characteristics. For example, many students enlist peer norms as justification for excluding and/or ignoring one or two targeted peers (No et al. 1999). Thus, the Korean cultural context is an ideal setting for testing the power of peer group norms. Second, the structure of schooling in South Korea consists of six years of elementary school (grades 1–6), three years of middle school (grades 7–9), and three years of high school (grades 10–12). Because instruction is classroom-based in both elementary and middle schools, the same classmates take nearly all their classes together during an academic year. This organizational structure indicates that the formation of peer relationships occurs more at the classroom level than at the school level. Korean middle school classrooms therefore were well-suited to test our multilevel hypotheses about defending behavior.

Current Study

The current study incorporated individual, social, and contextual and factors to investigate South Korean adolescents’ defending behavior in bullying situations. A large sample of more than 1,300 South Korean middle school students from 54 classrooms in 6 schools was recruited to examine how intrapersonal, interpersonal, and contextual factors concurrently predicted defending behaviors. In a multilevel framework, we first examined how intra- and interpersonal factors (e.g., gender, affective/cognitive empathy, perceived popularity) were associated with defending behaviors. Next, we assessed whether classroom-level prosocial norms as a contextual factor moderated the associations between intrapersonal factors, interpersonal factors, and defending behaviors.

Consistent with previous studies, we hypothesized that girls would be more likely to defend victims of bullying than boys would (e.g., Olweus and Endresen 1998; Rose and Rudolph 2006), and affective empathy, more so than cognitive empathy, would be positively associated with defending behaviors in bullying (e.g., Peets et al. 2015; Pöyhönen et al. 2010). For the interpersonal factor, we predicted that students who have a high social status in the peer group, as measured by high perceived popularity, would be more likely to defend victims (e.g., Peets et al. 2015). Finally, we anticipated that when anti-bullying attitudes (prosocial norms) are common in the classroom, students would be more likely to defend the victims of bullying. To our knowledge, no prior studies have examined how intrapersonal, interpersonal, and contextual factors interdependently affect defending behaviors among Asian middle school students.

Method

Participants

There were 1373 adolescents in this study (40% girls, M age: 14 yrs) selected from 54 classrooms in six coeducational middle schools in Seoul, South Korea. Data were collected in the spring semester of 2017. For this study, three grades were chosen from Korean middle schools (i.e., 34.3% first graders, 31.9% second graders, and 33.8% third graders), and the class sizes varied from 20 to 32 (M = 25 students).

Procedure

The data were collected through paper surveys. To ensure the precision of the translation, professional translators in Korea translated and back-translated the surveys from English to Korean and from Korean back to English. Surveys were administered during regular school hours in classrooms by the first author and two research assistants. Administering the survey took about 60 min. Only students who had written parental consent (approximately 98 %) were allowed to participate.

Measures

Defending behavior

Defending behaviors were measured with the Participant Role Questionnaire (α = .96; PRQ; Salmivalli and Voeten 2004). Students were presented with 15 items describing five participant roles in bullying situations, and they were asked to evaluate, on a 3-point scale (0 = never, 1 = sometimes, 2 = often) how often each of their classmates behaved in the ways described. The names of all classmates were printed on the questionnaire. The students evaluated the behavior of all classmates, except for themselves. Among five role scales, the defender scale consists of three items (i.e., “Tries to make others stop bullying;” “Comforts the victim or encourages him/her to tell the teacher about the bullying;” “Tells others to stop bullying or says that bullying is stupid”). Each student’s peer-evaluated sum score on each scale was divided by the number of classmates.

Empathy

We measured affective and cognitive empathy with adolescent self-reports on the empathic concern subscale (EC, α = .76; e.g., “I often have tender, concerned feelings for people less fortunate than me.”) and perspective taking subscale (PT, α = .80; e.g., “I try to look at everybody’s side of a disagreement before I make a decision.”) of the Interpersonal Reactivity Index (IRI; Davis 1980). Both subscales are composed of seven items scored on a 5-point rating scale (1 = does not describe me well, 5 = describes me very well). Scores for both subscales were averaged across the respective items, and higher scores reflect greater empathy.

Perceived popularity

As part of a peer nomination procedure, students were asked to list the names of the students in their classrooms who are the most popular and the least popular (e.g., Coie and Dodge 1983). Students were given a roster of all classmates arranged alphabetically by last name, and they were advised that they could list as many names as they wanted for these questions. The peer-nomination sum score of each item was standardized by classroom, and perceived popularity for each student was calculated by subtracting their standardized least popular nomination score from their standardized most popular nomination score.

Prosocial norms

Students’ pro-victim attitudes were measured by 10 items (α = .80; Salmivalli and Voeten 2004). Participants answered the extent to which they agree or disagree using a 5-point response scale ranging from 0 = strongly disagree to 4 = strongly agree (e.g., “One should try to help the bullied victims,” “Making friends with the bullied victim is the right thing to do”). Each student’s pro-victim attitude was calculated by averaging the students’ scores on 10 items. Then, a classroom-level indicator of pro-victim attitudes as prosocial norms was created by averaging the individual average score of pro-victim attitudes for each classroom.

Control variables

Two individual level measures were included as control variables. Earlier studies of peer nominations for popularity and related social reputations (e.g., Parkhurst and Hopmeyer 1998) recommended that researchers should not misidentify students perceived as popular with those perceived as being liked or aggressive. The perception of popularity status in peer groups among adolescents was based on evaluation of peers’ influence or dominance rather than just being liked (Eder 1985). To make the meaning of perceived popularity clear, we controlled for nominations of being liked and being aggressive. In the same peer nomination measure used to assess reputation as popular, students were asked to list the names of the other students in their classrooms whom they liked the most (“the people in your classroom you like the most”) and to list the classmates who “pick on other kids” (being aggressive). These variables were standardized the same way as the perceived popularity variable. In addition to two kinds of peer nominations, we also controlled for the student’s grade level in middle school.

Analytic Plan

We used Hierarchical Linear Modeling (HLM, Bryk and Raudenbush 1992; Raudenbush and Bryk 2002) to analyze our hypotheses concerning individual, reputational, and classroom level effects. Students were nested within 54 classrooms in our sample. HLM simultaneously assesses relationships within and between hierarchical levels and accounts for the shared variance in hierarchically structured data. This approach prevents both the incorrect partitioning of variance into variables and the increased risk of making a Type 1 error (Raudenbush and Bryk 2002). Missing data was handled via listwise deletion at the first level according to HLM protocol. Based on recommendations by Raudenbush and Bryk (2002), in order to compare the relative proportion explained by the addition of predictors, three models must be fit: 1) an unconditional model with no predictors, 2) a conditional model with only student level predictors and 3) a conditional model with the same student (Level 1) predictors as in Step 2 but with the addition of classroom (Level 2) predictors. Step 1 is used to calculate the intraclass correlation, the difference between Step 1 and Step 2 can be used to assess the relative proportion explained by all the student level predictors and the differences between Steps 2 and 3 are used to calculative the relative proportion explained by the classroom predictors.

Results

The means, standard deviations (SD), and correlation coefficients for student-level variables of main effects are represented in Table 1, separately by gender. Girls scored significantly higher than boys on defending behavior, affective empathy, and cognitive empathy, while there was no gender difference in popularity nominations by peers. All of these variables were significantly positively correlated with each other in the predicted direction for both boys and girls.

Table 1 Correlations, mean, and standard deviation among study variables for student-level models, separately by gender
Full size table

Three Steps of the in HLM Model

Step 1: The unconditional model

We first ran an unconditional hierarchical linear model (HLM) to assess the intraclass correlation (ICC) or the proportion of variance in defending behavior due to schools. To do so, we estimated the variance within classrooms σ2 as well as the variance between classrooms τ00. When we ran the model without any predictors on both Level-1 and Level-2, we obtained an estimated variance within classrooms σ2 = 0.028, and the variance between classrooms τ00 = 0.020. The ICC for Step 1 can be calculated as:

$$\frac{{\hat \tau _{00}\left( {Step\,1} \right)}}{{\hat \sigma ^2\left( {Step\,1} \right) + \hat \tau _{00}\left( {Step\,1} \right)}}{\mathrm{ = }}\frac{{0.02046}}{{0.02784{\mathrm{ + }}0.02046}}{\mathrm{ = }}0.424$$

The ICC was 0.424, indicating that 42% of the variance in defending behavior is attributed to between-classroom differences. In addition to the ICC, the chi-square test of between-classroom variance showed that statistically significant variance existed between classrooms in defending behavior: χ2(53) = 942.390, p < .001.

Step 2: The main effects model

To explore defending behavior as a function of student characteristics, we added gender, affective empathy, cognitive empathy, and perceived popularity as intrapersonal and interpersonal predictors to the unconditional model, controlling for grade level and nominations for being liked and being aggressive. We controlled for grade level because preliminary analyses showed no main effects or interactions involving this variable. Following Bryk and Raudenbush’s recommendations (1992), affective empathy, cognitive empathy, perceived popularity, likability, and aggression were centered around the group mean to yield more accurate estimates of the intercepts, while gender and grade were dummy coded (gender: male = 0; grade: two dummy variables first grade and second grade, while third grade was omitted to serve as the reference group).

The results from the estimation of the main effects model showed a significant effect of affective empathy and popularity on defending behavior, controlling for grade, likability, and aggression. As expected, affective empathy (b = 0.037, SE = 0.009, p < .001) and a reputation as popular (b = 0.222, SE = 0.036, p < .001) both had a positive effect on defending behavior; however, cognitive empathy was not significantly associated with defending behavior. In terms of gender differences, females were more likely to defend victims than males (b = 0.058, SE = 0.009, p < .001). Although we did not have a specific hypothesis about the interaction between student-level predictors, our results showed an interaction effect between gender and perceived popularity (b = 0.120, SE = .048, p < .05). As shown in Fig. 1, the association between popularity and defending behavior was stronger for girls than boys.

Fig. 1
figure 1

Associations between perceived popularity and defending behavior as a function of gender

Full size image

For the random effects model, we examined whether there was significant variability in slopes for each of the student-level predictors on defending behavior. None of the predictors had significant variability in slopes. Therefore, we removed random slopes from each of these remaining predictors in our third step. The student-level residual variance \(\hat \sigma ^2\left( {Step\,2} \right)\) at Step 2 decreased compared to the residual variance in the unconditional model at Step 1 \(\hat \sigma ^2\left( {Step\,1} \right)\). The proportion of variance explained at the student level can be calculated as:

$$\frac{{\hat \sigma ^2\left( {Step\,1} \right) - \hat \sigma ^2\left( {Step\,2} \right)}}{{\hat \sigma ^2\left( {Step\,1} \right)}}{\mathrm{ = }}\frac{{0.02784 - 0.02066}}{{0.02784}}{\mathrm{ = }}0.26$$

This means that 26% of the variability in defending behavior can be attributed to the full set of student level predictors entered at this step.

Step 3: The cross-level interaction model

In the final step, to estimate the classroom context effects due to differing levels of prosocial norms at the classroom level, we combined the student-level predictors, centered around the group mean, and the classroom-level predictor (prosocial norms), centered around the grand mean. After fitting a cross-level interaction model with classroom-level prosocial norms on all student-level predictors, we found that only perceived popularity had a significant cross-level interaction effect with classroom-level prosocial norms on defending behavior (b = 0.395, SE = 0.141, p < .01). Thus, all other cross-level interaction terms with prosocial norms were removed. The significant cross-level interaction involving perceived popular reputation and prosocial school norms is shown in Fig. 2. The values show the defending behavior rating for individuals who were either high (1 SD above the mean) or low (1 SD below the mean) on perceived popularity and were in classrooms with either high (1 SD above the mean) or low (1 SD below the mean) prosocial norms. The relationship between being perceived as popular and defending behavior was stronger when classroom-level prosocial norms were higher. That is, more popular students in classrooms with higher prosocial norms at the classroom-level were more likely to defend victims of bullying. In order to calculate the proportion of variation explained due to the addition of classroom predictors, a comparison must be made between Steps 2 and 3 given that both steps require the same set of student level predictors. This can be calculated as:

$$\frac{{\hat \tau _{00}\left( {Step\,2} \right) - \hat \tau _{00}\left( {Step\,3} \right)}}{{\hat \tau _{00}\left( {Step\,2} \right)}} = \frac{{0.02126 - 0.01416}}{{0.02126}}{\mathrm{ = }}0.33$$
Fig. 2
figure 2

Associations between perceived popularity and defending behavior as a function of school-level prosocial norms

Full size image

The proportion of variance explained by prosocial norms is 0.33, which means that adding prosocial norms as a classroom predictor reduced the classroom variability in outcomes by 33% (see Table 2).

Table 2 Fixed effects and random effects for the three-step hierarchical linear model
Full size table

Sensitivity Analyses

Given the organization of instruction in South Korea, the current study focused on classroom-level contextual factors. We collected data from 54 classrooms in six schools in the city of Seoul. Although classrooms are nested within schools, we chose not to model a three-level HLM because the number of schools to warrant analysis at the school level was not sufficient (Maas and Hox 2005). However, for exploratory purposes in this sensitivity analysis, we modeled a three-level rather than a two-level hierarchical structure with school as a level 3 variable.

First, we ran a fully unconditional model that did not include predictor variables to estimate the proportion of variation across the three different levels; (1) among students within classrooms, σ2, (2) between classrooms within schools, τπ, and (3) between schools, τβ. The proportion of variance of students within classrooms, between classrooms within schools, and between schools was 0.41, 0.55, and 0.03, respectively. The chi-square test showed that statistically significant variance exists between classrooms (χ2(47) = 704.53, p < .001) in defending behaviors. Moreover, there was marginally significant variance between schools (χ2(5) = 9.94, p = 0.076). These results support our belief that defending behavior in South Korean middle schools is probably best modeled at the classroom level due to the low proportion of variance at the school level.

Next we ran a fully conditional three-level model where we included the same predictors at the student- and classroom-level as in the original model but included aggregated school-level prosocial norms at Level 3. This model showed that perceived popularity had a significant cross-level interaction with school-level prosocial norms on defending behavior (b = 1.36, SE = 0.48, p < .01). That is, in schools with higher prosocial norms, popular students defended victims of bullying more, over and above the interaction of perceived popularity with classroom-level prosocial norms. Moreover, the rest of the predictors at the individual level were still significant controlling for school-level prosocial norms, indicating that these student-level predictors are robust to the three-level specification.

Discussion

For two decades, bullying has been conceptualized as a group process, involving multiple participant roles that take into account peer dynamics about power and status as they unfold in contexts most salient to youth. Acknowledging participant roles, many researchers agree that anti-bullying interventions should strengthen defending behavior and target the classroom or school level as well as the individual level (Salmivalli 2014). It is not enough to document individual characteristics of students that promote defending behavior without understanding how these person factors interact with the larger peer context. However, there are few empirical studies on the determinants of defending behavior with the needed multilevel approach, including those focusing on non-White participants. The research reported here examined defending behaviors of South Korean middle school students in bullying situations and how these behaviors were associated with intrapersonal, interpersonal, and contextual factors in a multilevel framework.

Consistent with previous studies on the intrapersonal factors associated with bullying (e.g., Caravita et al. 2009; Gini et al. 2008; Peets et al. 2015; Pozzoli and Gini 2010; Salmivalli et al. 1996), the present findings indicate that girls were significantly more likely than boys to defend victims of bullying. In addition, while cognitive empathy was found to have no associations with defending behaviors, affective empathy was positively associated with defending. Moreover, including interpersonal factors in the analysis showed that individuals who were considered popular by their peers had a greater tendency to defend victims in bullying situations. Since one of the research goals was to identify the effect of perceived popular status on defending behaviors, we controlled for likeability and aggression nominations in our analysis. Thus, the current study showed that perceived popular status more rigorously measured played an important role in defending behavior. Given the fact that defenders require a powerful social status to confront bullies or counteract future bullying, defending behavior is probably something different from general altruism or prosocial behavior.

Although interactions between intrapersonal and interpersonal factors were unexpected, the results indicated an interaction between gender and perceived popularity. Girls with a higher social status among their peers tended to defend victims of bullying more often than did boys with high social status. Previous studies focusing on white European participants (e.g., Finnish, Italian) failed to find this particular gender association. One possible explanation for our findings could relate to gender roles in South Korean culture. Despite significant and progressive modernization, many East Asian countries (e.g., China, Japan, South Korea) are still heavily influenced by Confucianism and people routinely face notable gender inequality, which is indoctrinated via parent–child socialization or school textbooks (Basu et al. 2017). Under such gender norms in East Asian societies, there is still a high expectation placed on girls to be obedient and engage in caring behaviors, compared to boys (Kang et al. 2016; Lindsey 2015). Accordingly, it is possible that South Korean girls with higher social status were encouraged to confront bullies and assume the role of defender, thereby caring for and protecting victims of bullying.

Regarding contextual factors, the results of the present study were consistent with those of previous studies (e.g., Kärnä et al. 2010; Peets et al. 2015; Pozzoli et al. 2012), indicating that classrooms differ in levels of defending behaviors, which could partly be explained by classroom prosocial norms. The association between perceived popularity and defending behavior was especially strengthened when classroom-level anti-bullying attitudes were higher. This interaction between classroom-level prosocial norms and popularity reputation in the peer group could be due to peer expectations for prosocial behaviors (Bukowski and Sippola 1996). The higher the level of prosocial norms in classrooms, the more students would expect perceived popular students to defend victims of bullying, and in turn the more likely the students with a reputation as popular might be to defend victims. In a similar manner, as gaining social acceptance and approval from peers becomes salient in adolescence, social rewards from peers play an increasingly important role in motivating certain behaviors (Wentzel and Brophy 2014). Accordingly, the higher the degree of anti-bullying attitudes in a classroom, the more likely it is that students will view defending behaviors positively, and the attainment or maintenance of high social status would be provided by peers as a form of social reward. Among adolescents for whom attaining or maintaining high social status is a critical developmental goal (Adler and Adler 1998), those who are already perceived as popular in classrooms with a high level of anti-bullying attitudes may engage in defending behaviors to maintain their high social status. We believe that peer expectations and status maintenance in the peer ecology may be important mediating mechanisms to explain defender behavior.

Limitations and Future Directions

Although we believe that the present study has many strengths, we also acknowledge its limitations. It is unclear from these cross-sectional data whether the participants engaged in defending behaviors because they were perceived as popular or they had achieved high social status among their peers because they defended victims in bullying situations. It could be that students acquire higher reputations in classrooms that promote prosocial norms because they, themselves, are more prosocial (i.e., their behaviors are aligned with the classroom norms). In this case, perceived popularity is partially shaped by the classroom’s norms rather than being independent of this contextual factor, as proposed in the present analysis. Further longitudinal studies are needed to fully examine the bidirectional relations between perceived popularity, prosocial norms in classrooms, and defending behavior, as well as other determinants of defending behavior.

A second limitation is our measure of defending behavior. Following Salmivalli and Voeten (2004), we created a reliable 3-item measure of general defending behavior. Yet that measure assessed both victim-oriented defending (i.e., comforts the victim) and bully-oriented defending behaviors (i.e., tells the others to stop bullying; tried to stop the bullying). As victim-oriented defending behaviors may carry fewer risks than bully-oriented defending behaviors, it is anticipated that each type has unique associations with the intrapersonal, interpersonal and contextual factors of bullying. Thus, future studies should consider the heterogeneity of defending behaviors in bullying situations.

A third limitation is our approach to high social status, a key factor in our analysis of what it takes to be an effective defender. We defined social status as having a reputation as popular among peers. Although we distinguished perceived popularity from likability and having a reputation as aggressive, our measure did not capture the multiple meanings of high social status during adolescence. Future studies need to consider other reputational measures of status such as coolness, dominance, and admiration (e.g., Bellmore et al. 2011; Closson 2009; Hawley 1999; Rodkin et al. 2006).

Lastly, since our data showed high variability in defending behavior between classrooms that varied in prosocial norms, it will be important to consider additional ecological factors that may vary at the classroom-level such as teacher behavior in bullying situations. Teachers can influence the peer ecology, including classroom prosocial norms, through their relationships with students and their instructional practices around grouping students (e.g., Farmer et al. 2011; Gest and Rodkin 2011). Findings on teachers’ role in preventing bullying in non-Asian contexts have been mixed, in part because teachers in U.S. and European settings are often unaware of bullying; it is a behavior that occurs most often in places where adult supervision is minimal, such as on playgrounds, in hallways, or in lunchrooms (e.g., Veenstra et al. 2014). In the U.S. context, moreover, students often are reluctant to report bullying because they question whether their teachers will actually come to their aid (Troop-Gordon 2015). Because adults tend to be more respected in Asian cultures due to the heavy Confucian influence (Hofstede 1980), Korean classroom teachers may be in a unique position to help victims of bullying (i.e., adopt a defender role). We believe that more research is needed on factors associated with the likelihood that teachers will intervene to address bullying.

Despite its limitations, this study contributes important knowledge to the relatively limited literature on peer-reported defending behaviors in bullying situations, especially among Asian adolescents. Furthermore, through a multilevel analysis, this study replicated the previous finding that contextual factors, such as prosocial norms, explain the variability between classrooms to a substantial extent, thereby emphasizing the need for a peer-ecological approach to the study of defending behaviors (Espelage and Swearer 2004).

The present findings also have implications for anti-bullying interventions. Considering that adolescents often misinterpret their peers’ attitudes based on peer behaviors (i.e., pluralistic ignorance; Prentice and Miller 1993), it is necessary to conduct group activities in the classroom to build a better understanding of the anti-bullying attitudes held by peer group members. Through such group activities, adolescents could determine whether their peers endorse anti-bullying attitudes, even if most do not engage in defending behaviors during bullying situations. Adolescents may also realize that if they engage in prosocial behaviors, such as defending behaviors, they could be positively evaluated and rewarded by their peers. To achieve these goals, adolescents’ affective empathy (i.e., understanding of the victim’s situation) and anti-bullying attitudes must be equally enhanced. Moreover, as indicated by the association found between perceived popularity and prosocial norms, students with high social status, who have a significant influence on their peers (Dijkstra et al. 2008), need to be made aware of the importance of their defender role in the classroom. They also must be taught more effective strategies for defending victims in bullying situations. Anti-bullying intervention programs should focus on identifying students with high social status and the greatest influence over their peers, as well as providing group activities that promote prosocial norms among all students in a classroom.

Implications for Bullying in Racially/Ethnically Diverse Contexts

Although our multilevel analyses were carried out in a racially homogeneous Korean school context, we believe that the findings can aid our understanding of bullying in the U.S. context where schools and classrooms are much more racially and ethnically diverse. Although studies that focus specifically on the ethnic context of bullying are limited, we do know that the numerical representation of different ethnic groups is a critical factor (see Graham and Echols 2018). Students who are members of numerical minority ethnic groups are particularly vulnerable to being the victims of bullying, whereas classroom and school diversity–more ethnic groups of relatively equal size–are a protective factor (Juvonen et al. 2017).

Extending these findings to the context of defending behavior, it would be important to examine how ethnicity might moderate the relations among intrapersonal, interpersonal, and contextual factors considered simultaneously in a multilevel framework. At the individual level, we should consider race/ethnicity of potential defenders as well as the numerical representation of their ethnic group. It is less likely that even the most empathic student will come to the aid of a victim if their group is a numerical minority. Regarding social status, numerical size will contribute to perceived popularity given well-established ingroup preferences, just as the group norms are likely to be more powerful when they are endorsed by members of one’s own racial/ethnic group (see Graham and Echols 2018). In other words, the processes that we examined here can contribute to a more nuanced understanding of the dynamics of defending behavior in a multiethnic context where the representation of different ethnic groups varies. A growing number of diversity indices are now available to model ethnic representation at the level of individuals as well as classroom and school contexts (Graham 2016). At the same time, most Asian countries, Korea included, are undergoing ethnic transformation with the influx of more immigrant populations (see Ha and Jang 2015). The ways in which U.S. researchers conceptualize ethnic diversity in bullying contexts can be a valuable starting point for thinking about defending behavior in a country of shifting numerical power balances. Thus programs of research on defending behavior in the U.S. and Korean contexts can be mutually beneficial.

Conclusion

Acknowledging that bullying is a group process involving multiple participant roles, anti-bullying programs often focus on encouraging bystanders to become defenders who are willing to come to the aid of victimized peers. Relatively little is known, however, about the characteristics of defenders, their motivations, or what contextual factors in the peer ecology support defending behavior. Taking a multilevel approach, we examined intrapersonal, interpersonal, and contextual factors associated with that behavior. At the individual level, students who were girls, had high affective empathy, and enjoyed a reputation as popular were more likely to defend victims. These personal and social assets were maximized when defenders resided in classrooms where the norms for prosocial behavior were high. Anti-bullying programs would do well to focus more on high-status and empathetic students as a route to changing peer norms about bullying. The study was conducted with a large sample of early adolescents in South Korea, an ideal cultural context for testing the intrapersonal, interpersonal, and classroom level factors examined here. Although Korea is still a racially homogeneous country, the findings can shed light on defender behavior in more racially and ethnically diverse societies. School bullying and efforts to reduce it are global challenges. Understanding what is universal and nation-specific to the social experiences of youth can greatly enrich our field.