Gender-specific or Common Classroom Norms? Examining the Contextual Moderators of the Risk for Victimization
The present study was funded by the Academy of Finland grant (project 114081) to the third author.
Abstract
We tested whether gender-specific vs. common classroom norms were more powerful moderators of the association between a risk factor (rejection) and peer victimization among girls and boys. The participants were 1220 elementary schoolchildren from grades 4–6 (with 10–13 years of age). We compared different multilevel models including combined vs. separate regressions for boys and girls, as well as the effects of norms of the whole class, same-sex classmates, and cross-sex classmates. Among girls, the association between rejection and victimization was strongest in classes where bullying behavior was common, and anti-bullying attitudes were rare among girls. Among boys, the strength of the slope of victimization on rejection could not be explained by either common or gender-specific classroom norms, but boys' level of bullying behavior was related to overall classroom level of victimization. The findings suggest that contextual factors may contribute to victimization especially among high-risk girls. The importance of exploring multiple levels of influence on children's social development is discussed.
Introduction
Peer victimization can be defined as children being systematically exposed to negative actions perpetrated by one or more of their peers. This can include both direct (e.g., hitting, verbal abuse) as well as more indirect behaviors (e.g., rumor spreading, social exclusion; Kochenderfer-Ladd & Ladd, 2001). It is associated with a vast array of adjustment difficulties, including anxiety, depression, loneliness, low self-esteem (for meta-analyses, see Card & Hodges, 2008; Hawker & Boulton, 2000), poor academic performance, dislike and avoidance of school (Buhs & Ladd, 2001; Kochenderfer & Ladd, 1996), and lack of friends (Hodges, Boivin, Vitaro, & Bukowski, 1999; Hodges & Perry, 1999; Pellegrini, Bartini, & Brooks, 1999). Given the abundance of adjustment difficulties associated with victimization, it is important that researchers not only look at risk factors for victimization, but also at environmental factors that temper or worsen risk for children who are vulnerable to victimization.
One of the most robust correlates of victimization is peer rejection (DeRosier & Thomas, 2003; Ladd & Troop-Gordon, 2003; Olweus, 1978; Perry, Kusel, & Perry, 1988; Salmivalli, Lagerspetz, Björkqvist, Österman, & Kaukiainen, 1996). A child is generally deemed rejected by their peers when majority of the peer group indicates they dislike the child (Coie, Dodge, & Copotelli, 1982). Rejection is not only a consequence, but also a risk factor for victimization: longitudinal findings have found it to be associated with increases in victimization over time (Hodges & Perry, 1999; Salmivalli & Isaacs, 2005). Rejected children may be perceived as legitimate targets for peer abuse because they are socially devalued. Peers tend to express negative attitudes toward rejected classmates, viewing them as more responsible for their transgressions, compared to well-liked peers (Hymel, 1986). This negative bias may signal to aggressive children that their attacks on rejected children may be warranted and will not result in censure from the peer group (Salmivalli, 2010; Salmivalli & Peets, 2009).
Although rejection is correlated with victimization, not all rejected children are necessarily victimized (Perry et al., 1988). The association between rejection and victimization varies significantly across classrooms (Kärnä, Voeten, Poskiparta, & Salmivalli, 2010; Sentse, Scholte, Salmivalli, & Voeten, 2007), and part of this variation can be explained by the degree to which classmates tend to reinforce the bully or support their victimized peers (Kärnä et al., 2010). The main purpose of the present study was to shed further light on conditions in which rejection is more or less likely to be associated with victimization. Studies on classroom norms (e.g., Henry et al. (2000; Salmivalli & Voeten, 2004) have so far focused on common classroom norms, assuming that attitudes and behaviors are more or less shared by all children in a class. We wanted to test whether the behaviors and attitudes of same-sex classmates, opposite-sex classmates, or classmates at large constitute a more important context weakening or strengthening the rejection–victimization link.
Protective and Risk-enhancing Environments
There is a growing need to explicate conditions or characteristics that might exacerbate risk or protect rejected children from victimization. Work evaluating person × environment interactions suggests that not only do individual factors affect development, they also work in concert with environmental conditions (e.g., Bates, Pettit, Dodge, & Ridge, 1998; Bellmore, Witkow, Graham, & Juvonen, 2004; Kochanska, 1995; Rothbart & Ahadi, 1994; Rubin, Burgess, Dwyer, & Hastings, 2003). For early adolescents, the peer group becomes an environment that has increasingly profound effects on development (Dishion, Patterson, Stoolmiller, & Skinner, 1991; Harris, 1995). There is evidence that certain risk factors for victimization are primarily associated with victimization in social contexts that disinhibit aggression or fail to protect at-risk children from aggressive attacks (e.g., when children lack friendships; Hodges et al., 1999; Hodges, Malone, & Perry, 1997; Hodges & Perry, 1999; Schwartz, Dodge, Pettit, Bates, & Conduct Problems Prevention Research Group, 2000; Schwartz, McFadyen-Ketchum, Dodge, Pettit, & Bates, 1999). Thus, these studies support the idea that interpersonal or contextual factors are pivotal in understanding when risk factors foster victimization.
Besides children's friendships or close peer networks (for review, see Card & Schwartz, 2009), the classroom might be an extremely relevant and powerful context for children's social development (Perry & Weinstein, 1998) because nearly all children spend a large portion of their day interacting with their classmates. Although interactions among classmates have been the focus for most peer relations research, studies on the influences of classroom-level factors related to bullying-related behaviors are less common.
The influence of classroom factors on at-risk children may rely on classroom norms (Salmivalli, 2010). Normative or high-frequency behaviors or attitudes can set the climate for acceptable and unacceptable behaviors in a class. Several studies have shown that engagement in behavior that is normative is associated with peer acceptance, even when the normative behavior, such as aggression, is not adult sanctioned (Boivin, Dodge, & Coie, 1995; Chang, 2004; Chen, Chang, & He, 2003; Ojala & Nesdale, 2004; Sentse et al., 2007; Stormshak et al., 1999; Wright, Giammarino, & Parad, 1986). Therefore, when bullying is normative in the classroom, the perpetrators of aggressive behavior may be more supported and accepted by classmates than in classrooms where this behavior is less normative (Salmivalli, Voeten, & Poskiparta, 2011). In addition, Henry et al. (2000) found that individual levels of aggression tended to decrease when children were in classes in which aggression was highly associated with rejection.
High levels of bullying among peers may not only increase the overall incidence of victimization (Hanish, Ryan, Martin, & Fabes, 2005) and contribute to increasing aggression among members of these aggressive cliques (Espelage, Holt, & Henkel, 2003), it may also place rejected children at heightened risk for victimization. In a classroom where bullying behavior is normative, rejected children may be especially enticing targets because of their devalued social status and low social clout: they are less likely to receive the sympathy and support that might occur in a classroom that largely does not condone bullying. Rejected children are also viewed as being more responsible for their transgressions than well-liked peers (Hymel, 1986); so when bullying is supported in a class, this negative bias may result in heightened risk for victimization for these vulnerable children. Bullying may be viewed as non-normative when there is a low frequency of bullying. In such classes, rejected children may not hold such a high value as potentially easy and attractive targets.
In addition to behavior, classroom attitudes may also form the basis of what is considered normative or acceptable and what is not. Pro-bullying attitudes have been found to be concurrently associated with actual bullying behavior (Henry et al., 2000; Pellegrini et al., 1999; Scholte, Sentse, & Granic, 2010) and predictive of increases in aggression over time (Huesmann & Guerra, 1997). Class levels of pro-bullying attitudes have also been found to predict increases in individual levels of aggression both directly and indirectly, through their influence on individual children's attitudes about aggression (Henry et al., 2000). In addition, classes characterized as being more approving of bullying behavior tend to have higher levels of bullying and victimization (Henry, 2001; Scholte et al., 2010). When many children in the classroom view bullying as innocuous, it may flourish within a classroom and make at-risk children particularly susceptible to victimization. When the classroom climate is characterized primarily by attitudes that oppose bullying, however, children may reap the protective benefits of a context that condemns bullying. Given that rejected kids are already at risk for victimization, being embedded in a context that supports bullying may exacerbate risk, whereas being in a classroom that opposes bullying may protect these at-risk children.
Although many of the aforementioned studies highlight the need to study the effects of contextual norms for aggression, they have primarily focused on aggressive behavior or on the acceptance (or rejection) of a child when the child engages in normative (or non-normative) behavior. There has been a dearth of studies examining how classroom norms affect victimization and, to our knowledge, only two studies have directly evaluated how known risk factors can translate into victimization depending on classroom norms. Hanish et al. (2005) found that same-sex classroom levels of aggression of preschool and kindergarten children were positively associated with victimization, but they failed to detect significant moderating effects of class level of aggression on the relation between children's social relationship quality (e.g., acceptance) and victimization. However, considerable power limitations were clearly acknowledged by the authors, suggesting the need to further examine the moderating role of classroom context factors utilizing substantially larger samples. Kärnä et al. (2010) examined the role that classroom bystander behavior played in either supporting or thwarting victimization for high-risk children. They found that social anxiety and peer rejection were most predictive of victimization when children were embedded in classrooms characterized by having many students who reinforced bullying behavior and few that defended the victim. However, the authors failed to examine whether these processes operate similarly for both boys and girls and did not investigate the unique effects of the behaviors of same- vs. opposite-sex classmates.
The Role of Gender in Differentiating the Influence of Context
The influence of context may differ for boys and girls. Males are often found to have more independent self-construals, meaning that they view others as being separate from the self and are largely guided by internal states, but females tend to be more interdependent, viewing others as part of the self and being guided by the need to maintain connectedness with important others (Cross & Madson, 1997). Girls have been characterized as being more relationship oriented than boys; whereas girls often have communal goals and select interpersonal styles that are focused on maximizing social harmony, boys tend to be more socially provocative and focused on more agentic goals that largely serve to meet their own needs (Fabes & Eisenberg, 1992; Kochenderfer-Ladd & Skinner, 2002; Maccoby & Jacklin, 1974; Ojanen, Grönroos, & Salmivalli, 2005; Rose & Asher, 1999; Rose & Rudolph, 2006). With regard to the use of aggressive behavior, girls tend to be more concerned with negative social repercussions but boys focus more on being able to control the victim through the use of aggression (Boldizar, Perry, & Perry, 1989). In classes where aggression is highly associated with peer disapproval, girls are less inclined to engage in aggression than boys (Henry, 2001). In addition, compared with boys, girls' behaviors in bullying situations are more influenced by contextual factors such as levels of bullying in their friendship networks or in their classroom; conversely, boys' behaviors in bullying situations are generally more influenced by personal factors (Salmivalli, Lappalainen, & Lagerspetz, 1998; Salmivalli & Voeten, 2004). Overall, normative behaviors in girls' friendship networks and classrooms tend to have a more profound effect on bullying behavior for girls, compared with boys. Due to the pivotal role that interpersonal relations play for girls, class norms may have greater effects on girls' behaviors. In addition, there is evidence that children tend to segregate into same-sex groups (Bukowski, Gauze, Hoza, & Newcomb, 1993; Maccoby, 1990), suggesting the potential importance of same-sex classmates in determining norms for children. Taken together, these studies highlight the importance of systematically evaluating gender-specific models when examining the effects of context on social behavior for children.
Aims and Hypotheses for the Present Study
The focus of the current study is on the contextual factors affecting the strength of the association between rejection and victimization. In line with Kärnä et al. (2010), it was hypothesized that the normativeness of bullying behavior in a classroom will moderate the association between rejection and victimization, such that rejection will be most strongly associated with victimization in classrooms where bullying behaviors are occurring at high levels or where anti-bullying attitudes are rare. More importantly, we examined potential gender-specific processes due to children's tendency to segregate into same-sex groups and due to the pivotal role that social context plays especially on girls' social behavior (e.g., Henry, 2001; Maccoby, 1990; Salmivalli et al., 1998; Salmivalli & Voeten, 2004). This was done by testing competing cross-level models to find out whether gender-specific vs. common classroom norms are more powerful moderators of the slope of victimization on rejection. It was hypothesized that norms among same-sex classmates will be most influential and that the effects of classroom context will be most pronounced in girls.
In addition to the earlier work on the effects of peer socialization (e.g., antisocial behavior; Dishion, Spracklen, Andrews, & Patterson, 1996; Patterson, 1993), more recent work has found support for the influence of classroom norms largely in the period from late childhood though adolescence (e.g., Dijkstra, Lindenberg, & Veenstra, 2008; Kärnä et al., 2010; Salmivalli & Voeten, 2004; Sentse et al., 2007). Therefore, we focus on the period of late childhood/early adolescence to investigate the effects of classroom context.
Method
Participants
Participants were 1220 elementary schoolchildren (600 girls) from grades 4–6 (aged 10–13). They represented 48 classes in 16 schools located in southern Finland. Class sizes varied from 19 to 33 with a mean class size of 25 students. Classes tended to have roughly equal percentages of girls and boys, with most classes being composed of 40–60 percent girls (total range 29–65 percent girls). Elementary schoolchildren in Finland tend to spend the whole school day with their class, and often children are familiar with their classmates from previous years of being in the same school, often in the same class. Although no information was available concerning individual socioeconomic status levels, due to the Finnish comprehensive school system, the students represented all social classes, with no notable between-school socioeconomic differences.
Procedures
Data collection took place in the fall, 2 months after the start of the school year. Consistent with the Finnish Human Subjects Protection regulations, passive parental consent procedures were used after obtaining approval for the study by the school district superintendent and the principals of the schools. Parents received an information letter from the investigators explaining the goal of the study and the procedures involved, including the phone number of the principal investigator (the third author). Parents were specifically instructed to sign a form letting the teacher know if they wished their child not to take part in the study.
Data were collected in classrooms during regular school hours. The instructions for all questionnaires were read aloud by two trained research assistants, and the confidentiality of responses was emphasized to the students. Children completed the questionnaires independently at their desks, under the supervision of the two research assistants.
Measures
Definition of Bullying. The following definition of bullying (based on Olweus, 1996) was read to the students and was written on the questionnaires:
It is bullying when one child is repeatedly exposed to harassment and attacks from one or several other children. Harassment and attacks may be, for example, shoving or hitting the other one, calling him/her names or making jokes about him/her, leaving him/her out of the group, taking his/her things or any other behavior meant to hurt the other one. It is not bullying when two students with equal strength or equal power have a fight, or when someone is occasionally teased, but it is bullying when the feelings of one and the same student are intentionally and repeatedly hurt.
Victimization. Victimization was assessed by presenting children with the definition of bullying and then asking them to write down the names of children in their class who were being bullied. Victimization scores were calculated by taking the proportion of nominations received out of the total possible nominations, thus resulting in possible scores ranging from 0 to 1. The validity in assessing victimization through peer nominations is supported by moderate to strong correlations found with teacher and self-reports of victimization (Perry et al., 1988). Peer nominations also tend to be markedly reliable, even single-item scales, because they rely on aggregating across multiple respondents (Coie, Dodge, & Kupersmidt, 1990).
Bullying. This scale was used to assess the degree to which children engaged in active bullying behaviors, acting as the bullying ‘ringleader’. After being presented with the definition of bullying, students rated every child in their class (0 = never, 1 = sometimes, 2 = often) on three questions about their bullying behaviors during bullying situations (i.e., starts bullying, makes others join in the bullying, always finds new ways of harassing the victim). Person means were calculated for each item by dividing the weighted number of nominations received by the total number of potential nominators. These item means were then averaged across the three items, producing scale scores that could range from 0 to 2. The bullying scale was highly internally consistent, with a Cronbach's alpha coefficient of .93, and has been found to correspond with self-reports (Salmivalli et al., 1996) and teacher reports of bullying (Salmivalli & Nieminen, 2002).
Anti-bullying Attitudes. Students' attitudes about bullying were assessed by a 10-item self-report scale. Students were asked to evaluate how strongly they agreed or disagreed with statements about bullying (0 = strongly disagree, 4 = strongly agree). Some sample items are ‘bullying is stupid’, ‘it is the victim's own fault that they are bullied’ (reverse coded), and ‘one should report bullying to the teacher’ (see Salmivalli & Voeten, 2004, for the full scale). Scale scores were computed by averaging the scores of each item. Higher scores on this scale reflected stronger attitudes against bullying and lower scores reflected more pro-bullying attitudes. Data were obtained for 1134 students; 7 percent of the students had a missing value. The scale demonstrated adequate reliability (α= .75) and has shown to be predictive of bullying-related behaviors (bullying others, assisting the bully, and reinforcing the bully), while being negatively associated with supporting the victimized peers (Salmivalli & Voeten, 2004).
Rejection. Children were asked to write down the names of the three classmates whom they liked least. Rejection scores were calculated by taking the proportion of liked least nominations received out of the total possible nominations, thus resulting in possible scores ranging from 0 to 1.
Results
Analysis Strategy
Multilevel modeling was used because we had nested data and explanatory variables at two levels. Students were nested within classrooms, which may result in non-independence of observations; children from the same class are more likely to be similar than children from different classes (Julian, 2001). Multilevel modeling allows the researcher to decompose the variance into variance associated with individual students (within-class variability) and variance associated with the social context of the classroom (between-class variability). Therefore, associations between variables can be accurately evaluated at the level of the individual, as well as at the level of the class. We used explanatory variables at both the student level and the classroom level. Traditional methods produce standard errors that are inaccurate when the nested structure of classroom-based data is ignored (Bliese & Hanges, 2004; Julian, 2001; Lee, 2000; Raudenbush & Bryk, 2002) and treats dependency in the data as a nuisance rather than as a factor to be modeled.
Results are presented in three sections. Firstly, within-class and between-class correlations are presented. Secondly, we provide intra-class correlations (ICCs) for each variable, which indicate the proportion of variance between classes. In the third section, we present multilevel models to test cross-level interactions, that is, to test moderator effects of classroom-level variables. These models were examined by first testing whether the (gender-specific) slopes of victimization on rejection varied between classes. Then we tested cross-level interactions examining whether the slopes of victimization on rejection varied as a function of classroom levels of bullying and anti-bullying attitudes. For the cross-level interactions, comparisons were made among various models in which we explored combined vs. separate regressions for boys and girls, as well as the effects of norms of the whole class, same-sex classmates, and cross-sex classmates.
The models were estimated by full maximum likelihood,1 using Mplus Version 3 (Muthén & Muthén, 2004). Most variables were considerably skewed and were therefore transformed to normal scores, utilizing the rankit procedure, prior to conducting all analyses. Predictor variables were scaled such that the score 0 indicated an absence of risk-enhancing behavior (rejection, bullying). The scale for anti-bullying attitudes was grand-mean centered. Norms variables at the classroom level were obtained by aggregating individual-level variables to the classroom level through averaging. In the current study we examine two types of aggressive norms. The first are normative levels of aggressive behavior, which reflect the average level of aggressive behavior in the classroom, what Henry et al. (2000) described as descriptive class norms. The second type are normative attitudes about aggression, which reflect the classes' average belief in the acceptability of aggressive behavior, referred to by Henry and colleagues as injunctive norms. To examine gender-specific norms means are examined separately for the boys and the girls in a classroom.
Correlations
Correlations between variables were examined utilizing a two-level analysis in Mplus (Muthén & Muthén, 2004). Table 1 shows the pooled within-classroom correlation matrix and the between-classroom correlation matrix for the whole sample, and separately by gender.
| 1 | 2 | 3 | 4 | |
|---|---|---|---|---|
| Whole sample | ||||
| 1. Rejection | — | |||
| 2. Victimization | .44*** | — | .53*** | −.03 |
| 3. Bullying | .38*** | .04 | — | −.19 |
| 4. Anti-bullying attitudes | −.12*** | .03 | −.28*** | — |
| Girls only | ||||
| 1. Rejection | — | |||
| 2. Victimization | .44*** | — | .74*** | −.53*** |
| 3. Bullying | .28*** | .06 | — | −.25 |
| 4. Anti-bullying attitudes | −.06 | .06 | −.08 | — |
| Boys only | ||||
| 1. Rejection | — | |||
| 2. Victimization | .41*** | — | .54*** | .10 |
| 3. Bullying | .37*** | −.09* | — | −.24 |
| 4. Anti-bullying attitudes | −.08 | .07 | −.22*** | — |
- Note. Correlations presented below the diagonal represent individual-level correlations within classrooms. Correlations presented above the diagonal represent between-classroom correlations involving norms variables. All correlations were estimated using two-level analyses with Mplus.
- Ns for correlations below the diagonal were 1134 for the whole sample, 552 for girls, and 582 for boys. Missing values for the attitude scale reduced the total N by 7%. Ns for all correlations above the diagonal are equal to the number of classrooms, 48.
- * p < .05, ** p < .01, *** p < .001.
Within-classroom correlations revealed strong positive associations between rejection and victimization. The pattern of correlations among bullying, anti-bullying attitudes, and victimization differed somewhat at the individual (within classroom) and classroom levels. Within classrooms, individuals who bullied more tended to have more pro-bullying attitudes, but this association was most evident for boys. At the classroom level, however, this correlation was not statistically significant due to the low N. On the other hand, bullying and victimization were hardly correlated at the individual level within classrooms but showed a strong positive correlation at the classroom level.
Degree of Between-classroom Variability in the Study Variables
All variables were analyzed to evaluate the proportion of variance that could be explained due to individual differences (within classrooms) and classroom differences (between-classroom means).2 For victimization, there was statistically significant variability between classes, s2= .01, z= 1.69, p < .05. The ICC was .02, indicating that 2 percent of the total variance in victimization was due to similarity between classmates, the rest being due to individual differences. Rejection, however, had statistically significant variation only between individuals within classrooms, ICC = .00. Therefore, nearly all of the variability in rejection can be attributed to individual differences rather than clustering within classrooms.
Children's bullying behavior and anti-bullying attitudes were conceptualized as moderators at the classroom level and both indeed demonstrated statistically significant variability between classrooms. Fifteen percent of the variance in bullying (s2= .13, z= 3.87, p < .001; ICC = .15), and four percent of the variance in anti-bullying attitudes (s2= .04, z= 2.67, p < .01; ICC = .04) was due to between-classroom differences.
Between-classroom Moderators of Risk for Victimization
In order to examine the role of classroom-level factors (i.e., normative bullying behavior and anti-bullying attitudes) as moderators of the association between rejection and victimization, the first step, before examining gender-specific processes, was to fit whole-classroom models without distinguishing between norms for boys and girls. We started by testing whether the slope of victimization on rejection varied between classes, using gender and grade level as control variables. As expected, there was a statistically significant classroom-level variance in the slopes (s2= .02, z= 2.18, p < .05), meaning that the relationship between victimization and peer rejection differed between classrooms.
Next we constructed two models for testing the effects of bullying behaviors and of anti-bullying attitudes. In the first case, the model included peer-reported bullying behavior as a predictor at the individual level. In addition, the classroom mean of bullying behavior was included as a classroom-level predictor for the intercept of victimization (i.e., the mean level of victimization in a classroom) and for the slope of victimization on rejection. The latter represents a cross-level interaction, to assess the moderator effect of classroom norms. Similarly, a model was constructed using anti-bullying attitudes and their classroom means as predictors instead of bullying behavior. Our main interest was in the cross-level interactions, because these represent the effect of the classroom context on the relation between rejection and victimization. The results for the two models are shown in Table 2. The table shows for each predictor the unstandardized regression coefficient (Est.), its standard error (SE), and the z test. Firstly, the individual-level predictors are presented (for individual differences within classrooms), and next the classroom-level predictors (for differences between classrooms). Finally, the table includes the estimated variance components at the individual level (within-classrooms residual variance of victimization) and at the classroom level (intercept variance and slope variance). The classroom means of bullying behavior and anti-bullying attitudes are in the table under the name ‘class norms’. We were primarily interested in the classroom-level effects.
| Effects | Moderators | |||||
|---|---|---|---|---|---|---|
| Bullying behavior | Anti-bullying attitudes | |||||
| Est. | SE | z | Est. | SE | z | |
| Within classrooms | ||||||
| Intercept | −.32 | .13 | −2.46 | −.25 | .07 | −3.55 |
| Boy | .14 | .04 | 3.31 | .07 | .04 | 1.63 |
| Rejection | .01 | .01 | 1.71 | .35 | .03 | 11.89 |
| Bullying/attitudes | −.15 | .03 | −5.45 | .07 | .02 | 3.35 |
| Between classrooms | ||||||
| Grade levela | −.04 | .03 | −1.35 | −.04 | .03 | −1.37 |
| Class normsb | .13 | .09 | 1.52 | .01 | .14 | .06 |
| Class norms × rejection | .08 | .07 | 1.17 | −.08 | .10 | −.74 |
| Residual variances | ||||||
| Within classrooms | .39 | .02 | 23.79 | .40 | .02 | 22.86 |
| Classroom intercepts | .02 | .01 | 1.71 | .02 | .01 | 1.94 |
| Classroom slopesc | .01 | .01 | 2.02 | .02 | .01 | 2.26 |
- a Grade level scored as 0 = grade 4, 1 = grade 5, and 2 = grade 6. b Classroom means of respectively bullying behavior and anti-bullying attitudes. c Covariance between intercepts and slopes omitted.
Table 2 shows that no statistically significant relationships were found for the whole-classroom norms; most importantly, the classroom norms were not significantly related with the slopes of rejection (b= .08, p > .05 when using bullying behavior, or b=−.08, p > .05 when using anti-bullying attitudes). The slope differences between classrooms thus could not be explained by whole-classroom norms. At the individual level, gender appeared to have an effect on victimization when controlling for bullying behavior. The dummy variable ‘boy’ shows that boys are victimized more than girls after holding constant rejection and bullying behavior (b= .14, p < .01). This gender effect, however, disappeared in the model with anti-bullying attitudes.
The second step in our analyses was to investigate gender-specific processes to test our hypothesis that gender-specific norms are more powerful moderators of the relationship between victimization and rejection than whole-classroom norms are. Therefore, the slopes of victimization on rejection were evaluated separately for boys and girls. In the within part of the model, two random slopes for victimization on rejection were defined, one for boys and one for girls. The gender-specific slopes were made possible through multiplying the predictor, rejection, by appropriate dummy variables, one with the code 1 for girls and one with the code 1 for boys. See the Appendix for further information on the model specification. As was the case in the previous analysis for the whole classroom, both boys and girls showed statistically significant between-group variation in the slopes (girls: s2= .03, z= 2.13, p < .05; boys: s2= .02, z= 2.29, p < .05).
In all further two-level analyses we start from a model that designates separate slopes for girls and boys, utilizing only same-sex peers to calculate measures of classroom norms. Estimation of separate slopes permits examination of potential gender-specific moderator effects of classroom-level variables on these slopes. Cross-level interactions thus may differ for boys and girls. This model is then compared with models that specify common slopes for boys and girls (i.e., two gender-specific slopes that are constrained to be equal), models using the whole class (both boys and girls) to calculate measures of classroom norms (these models are already in Table 2), and models that include the influence of cross-sex class norms on the gender-specific slopes.
In the first model, classroom means of bullying behaviors were computed for girls and boys separately and entered as predictors of classroom levels of victimization3 and also of the gender-specific slopes of victimization on rejection to test cross-level interactions. The results of this model are presented in Table 3 together with the results of a second model in which the effects of same-sex classroom means of anti-bullying attitudes were examined. This table is similar to Table 2 except (1) there are now two random slopes for peer rejection, one for boys and one for girls, and consequently (2) two cross-level interactions, using (3) same-sex norms rather than whole-classroom norms. These models allow for gender-specific processes in two ways. Firstly, gender-specific norms (same-sex and cross-sex) were used instead of whole-classroom norms. Secondly, the effects of the norms on the risk for victimization, operationalized by the slope of victimization on peer rejection, were allowed to differ for boys and girls. Note, that the models had random intercepts, allowing for classroom differences in the level of victimization, but the gender effect on victimization was assumed to be constant across classrooms (see the Appendix for an explanation).
| Effects | Moderators | |||||
|---|---|---|---|---|---|---|
| Bullying behavior | Anti-bullying attitudes | |||||
| Est. | SE | z | Est. | SE | z | |
| Within classrooms | ||||||
| Intercept | −.37 | .14 | −2.71 | −.25 | .07 | −3.55 |
| Boy | .14 | .06 | 2.43 | .06 | .06 | 1.10 |
| Rejection for girls | .14 | .08 | 1.64 | .44 | .06 | 7.65 |
| Rejection for boys | .36 | .13 | 2.84 | .34 | .04 | 7.87 |
| Bullying/attitudes | −.16 | .03 | −5.83 | .08 | .02 | 3.79 |
| Between classrooms | ||||||
| Grade levelb | −.05 | .03 | −1.70 | −.05 | .03 | −1.53 |
| Girlclassc | −.10 | .17 | −.59 | .09 | .25 | .34 |
| Boyclassc | .34 | .15 | 2.18 | .03 | .15 | .18 |
| Girlclass × Rejection for girls | .56 | .16 | 3.42 | −.80 | .29 | −2.78 |
| Boyclass × Rejection for boys | .05 | .14 | .33 | −.07 | .16 | −.47 |
| Residual variances | ||||||
| Within classrooms | .37 | .02 | 23.24 | .38 | .02 | 22.32 |
| Classroom intercepts | .02 | .01 | 1.74 | .03 | .01 | 2.04 |
| Classroom slopes for girls | .02 | .01 | 1.79 | .03 | .02 | 1.87 |
| Classroom slopes for boys | .02 | .01 | 2.21 | .03 | .01 | 2.53 |
- a See the Appendix for further information on model specification. b Grade level scored as 0 = grade 4, 1 = grade 5, and 2 = grade 6. c Girlclass and Boyclass refer to respective same-sex class norms for the moderator variables.
As shown in Table 3, classroom levels of victimization were significantly predicted by classroom levels of bullying among boys, b= .34, p < .05, but not by bullying among girls, b=−.10, p > .05. When a classroom had a higher level of bullying by boys, then also a higher level of victimization tended to be present in the classroom as a whole. Due to the classroom variation in the coefficients for rejection, this result only reflects children with the score 0 on rejection (i.e., children that were not rejected by their peers, which is true for a vast majority of the children). The classroom variation in slopes of victimization on rejection for boys could not be explained by the level of bullying behavior by boys in a classroom, b= .05, p > .05. For girls, however, the level of bullying by girls significantly predicted variation in the slope of victimization on rejection, b= .56, p < .001. Follow-up analyses followed the procedures proposed by Aiken and West (1991) to examine the relation between a predictor and a criterion estimated as an unstandardized regression coefficient (b) at low (mean − 1 SD), medium (the mean), and high (mean + 1 SD) levels of a moderator. For girls, the positive relation of rejection with victimization became stronger when the girls in the class engaged in more bullying (see Figure 1). When there was no bullying at all among girls in a classroom, victimization was not significantly (p > .05) related with rejection; see the coefficient of rejection for girls in the first model of Table 3. At all three levels of bullying by girls depicted in Figure 1, however, there is a statistically significant positive relation of rejection with victimization.
Regression of victimization on rejection at different values of class bullying for girls.
We compared the model from Table 3 containing gender-specific regressions and gender-specific cross-level interactions with the common-regression model, which had only one cross-level interaction, restricted to be the same for boys and girls. We used the chi-square difference test and the sample-size adjusted Bayesian information criterion (BIC) to compare models (Sclove, 1987). A lower BIC value indicates a better fit. The gender-specific model fit clearly better, χ2dif (5) = 324.48, p < .001. With the common regression for girls and boys, the cross-level interaction was not statistically significant (b= .09, z= 1.18, p > .05). Furthermore, the fit of the common-regression model deteriorated when classroom means of bullying were used in lieu of same-sex means (BIC = 2403.83 vs. 2148.38),4 again pointing to the importance of considering the gender context rather than the whole classroom. This seemed the more important because, as noted before, the whole-classroom norms were not significantly related with the intercept nor with the slope of rejection (see Table 2). Next, we performed a more specific test of the hypothesis that the cross-level interaction differs for girls and boys. We computed separate gender-specific regressions but put a constraint on the model by forcing the cross-level interaction for girls to be equal to the cross-level interaction for boys, still using same-sex classroom means. The constrained model differed significantly from the unconstrained model, χ2dif (1) = 7.04, p < .01. We conclude that not only was it optimal to use same-sex classmates to measure classroom norms, but also the interaction effect of classroom bullying and rejection on victimization differed for girls and boys. A significant positive interaction effect was found for girls, whereas for boys the interaction effect was not statistically significant.
Finally, we addressed the issue whether the slope of victimization on rejection for girls was not only associated with girls' bullying, but also with classroom bullying by boys. The effect of boys' means on the slope for girls was not statistically significant, b= .20, z= .90, p > .05. Likewise, the effect of girls' bullying on the slope for boys was not significant, b=−.22, z=−1.24, p > .05. Adding the effects of cross-sex means did not improve the fit of the model, χ2dif (2) = 1.15, p > .05. In fact, the extended model fit worse than the original model: BIC = 1849.11 vs. 1843.54 for the original model. We may conclude that, for girls, primarily girls' bullying behavior in the classroom affects the slope of victimization on rejection. Neither girls' nor boys' bullying behaviors in the class significantly influenced the slope for boys.
Utilizing the same procedure, the effects of same-sex between-classroom variations in anti-bullying attitudes were explored; see the second model in Table 3. Classroom levels of victimization were not significantly associated with anti-bullying attitudes among girls, or among boys in the classroom. Also the slope of victimization on rejection for boys was not significantly associated with the class levels of anti-bullying attitudes among boys, b=−.07, p > .05. For girls, however, the classroom level of anti-bullying attitudes significantly qualified the effect of rejection on victimization, b=−.80, z=−2.78, p < .01, weakening the association in classes with higher levels of anti-bullying attitudes (see Figure 2). There was a positive and statistically significant association between rejection and victimization for girls in classrooms with a low or medium level of anti-bullying attitudes (at the medium level b= .21, z= 3.22, p < .01), but the relation was not statistically significant at high levels (+1 SD) of anti-bullying attitudes.
Regression of victimization on rejection at different values of class anti-bullying attitudes for girls.
As with bullying behavior, the model with gender-specific regressions fit clearly better than the common-regression model utilizing only one cross-level interaction that was restricted to be the same for both boys and girls, χ2dif (5) = 1233.96, p < .001. With common regressions for boys and girls, no statistically significant cross-level interaction could be detected (b=−.09, z=−.83, p > .05). Furthermore, when using a model with the means of whole classrooms in lieu of same-sex means, the BIC increased from1977.01 to 2273.16, and again in the model with whole-classroom norms, no statistically significant effects of the classroom norms were obtained (see Table 2). The specific test of equality of the cross-level interaction for girls and boys (using two separate, gender-specific, cross-level interactions constrained to be equal) yielded a statistically significant chi-square difference, χ2dif (1) = 5.22, p < .05, indicating that the cross-level interactions differ for girls and boys. Taken together, this pattern of results once again supports the utility of examining gender-specific processes within the classroom.
When the boys' means were added to the original gender-specific model as a predictor of the slope for girls, no statistically significant effect was obtained, b=−.09, z=−.41, p > .05. The effect of girls' anti-bullying attitudes on the slope for boys was also not significant, b= .43, z= 1.49, p > .05. Adding the cross-sex means resulted in a poorer fitting model χ2dif (2) = 2.33, p > .05; BIC = 767.72 vs. 762.33 for the original model. We may conclude that it is primarily anti-bullying attitudes among girls that affect the slope of victimization on rejection for girls. Again, no such contextual effects were found for boys.
In summary, for girls, having a high level of bullying in the classroom placed vulnerable girls (those higher on rejection) at greater risk for victimization. Classroom levels of anti-bullying attitudes among girls, however, had the reverse effect for girls. They helped buffer rejected girls from victimization. For boys, their level of bullying behavior in the class predicted overall class levels of victimization, but classroom-level factors were not associated with the slope of victimization on rejection.
Discussion
Classroom norms concerning bullying were shown to moderate the association between rejection and victimization, but this moderator effect was only found when using same-sex class norms and it was found for girls only. For boys, the association between rejection and victimization was not contingent on any of the classroom level norms. For girls, rejection was most strongly associated with victimization in classes with girls that had higher levels of bullying and pro-bullying attitudes. The relation between rejection and victimization was weakest in classes with girls that had lower levels of bullying and were attitudinally more opposed to bullying. Against expectations, no significant cross-level interaction was found for whole-classroom norms, like that found in the Kärnä et al. (2010) study. There was, however, a positive effect of descriptive class norms on the level of victimization in classrooms. As expected, higher levels of victimization were observed in classrooms with higher levels of bullying. The multilevel models show that the use of same-sex class norms and gender-specific moderator effects of these norms give better fitting models than the use of whole-classroom norms. The failure to detect statistically significant moderator effects of whole-classroom norms further underlines the importance of considering same-sex norms. Also, the analyses revealed that only same-sex norms and not cross-sex norms were associated with the risk of victimization.
The reason why context had a more powerful effect for girls may be due to their heightened concern about social consequences for their behavior (Boldizar et al., 1989). Girls may therefore search their environment for cues about the appropriateness of their behavior. Bullying (Nansel et al., 2001; Scheithauer, Hayer, Petermann, & Jugert, 2006; Solberg, Olweus, & Endresen, 2007) and overt aggressive acts (see Card, Stucky, Sawalani, & Little, 2008 for meta-analysis; Crick, Casas, & Mosher, 1997; Crick & Grotpeter, 1995; Rys & Bear, 1997) tend to be less frequent among girls; so bullying episodes may be more carefully premeditated, basing decisions on both the characteristics of the potential target and the classroom norms for appropriate behavior. Boys, for whom overt aggression is more normative, may look primarily at the characteristics of the potential target (e.g., is the child easy to dominate) or base their current behavior on their own previous bullying behavior (Salmivalli et al., 1998), ignoring the classroom climate. There is also some evidence that girls are more susceptible to peer influence (Schulenberg et al., 1999; for conflicting evidence, see Steinberg & Monahan, 2007) and that classmates tend to have negative biases toward rejected children (Hymel, 1986; Schuster, 2001). For girls, the felt pressure to bully socially disliked children may be quite salient in classes with norms that favor bullying. Among girls, rejected children may be seen as deserving of peer abuse and may not have the social support to ward off bullies, especially in classes where bullying is more normative (i.e., high levels of bullying behavior and pro-bullying attitudes). Female bullies may specifically choose rejected children as targets because the imbalance of power that is inherent in the definition of bullying may be easier to achieve when the target is socially rejected and therefore lacks the social power and stature afforded to more popular children. Also, given females' susceptibility to peer influence, peers may also sway girls not to bully these ‘easy targets’ when the bullying is non-normative.
Social harmony and cohesion are also very important for girls (Miller, Danaher, & Forbes, 1986; Ojanen et al., 2005). Rejected girls may be seen as being in the ‘out-group’. Therefore, in classes with norms that support bullying for girls, rejected girls may be especially targeted to support the cohesion of the rest of the group. Overall, for girls, being disliked may be a stronger predictor of victimization in a class in which bullying behavior and attitudes are normative, due to the powerful socialization processes found among girls. In addition, it was the norms among female classmates, rather than male classmates, which impacted the association between rejection and victimization. Therefore, although most children are grouped in mixed-sex classes, for girls, it may be the same-sex classmates that cast the most influential norms.
What was normative among boys, however, influenced the level of victimization in the whole classroom. Recent studies utilizing dyadic data show that victimization does not only occur between same-sex peers, but crosses sex boundaries as well (Rodkin & Berger, 2008; Veenstra, Lindenberg, Munniksma, & Dijkstra, 2010). Boys are more likely to choose girls as their targets rather than the other way around (Olweus, 2009). This might explain why bullying behavior among boys in a classroom increases the likelihood of both girls and boys in that class being victimized.
The current investigation was one of the few studies that examined classroom-level moderators of a known risk factor for victimization using a multilevel approach. This is an important step forward conceptually because bullying episodes can be seen as a group process (O'Connell, Pepler, & Craig, 1999; Salmivalli et al., 1996). The examination of risk factors in isolation from environmental conditions ignores the social interconnectedness that is inherent to peer relations. In addition, a multilevel approach allows for the exploration of cross-level interactions; individual-level associations can be introduced at the level of the group and other group-level variables can be probed for their influence on the association. A new feature in our multilevel models is the differentiation of the classroom context into gender-specific social contexts (cf. Chang, 2004) in a way that allowed us to test gender-specific cross-level interactions. Overall, multilevel approaches may help to produce more intricate and accurate models of social development and influence.
Implications for Interventions
Building upon the work of Kärnä et al. (2010), the results from this study point to the need to account for contextual factors when designing intervention and prevention programs. The ‘one-size-fits-all’ approach to applying generic interventions is likely to not have the most widespread efficacy. There is evidence to show that certain interventions are more or less effective depending on contextual factors such as peer, teacher, and school characteristics (Kallestad & Olweus, 2003; Metropolitan Area Child Study Research Group, 2002; Poulin, Dishion, & Burraston, 2001). Our study contributes to the growing body of literature showing multiple levels of influence on child adjustment. Multilevel conceptualizations and evaluation of interventions are vastly needed (Choi, 2003). Results from this study suggest that classroom factors may be central to designing tailor-made interventions. For example, targeting individual risk factors may be less fruitful in classes with girls that endorse pro-bullying attitudes.
In addition, caution may need to be taken when creating intervention groups given the evidence of the iatrogenic effects that are associated with aggregating antisocial youths (Dishion & Andrews, 1995; Dishion, McCord, & Poulin, 1999), which tend to be especially deleterious among the lower risk group members (Poulin et al., 2001). In our study we found that, for girls, high levels of bullying behavior in the class were associated with increased risk of victimization for rejected children. These effects may become even more pronounced if children who tend to bully infrequently gradually are shaped to bully socially marginalized children more consistently.
Limitations and Future Research
One limitation of the current investigation was the concurrent design. The fact that rejection and victimization, as well as all of our moderating variables, were examined contemporaneously precludes us from examining processes that lead to changes in victimization over time. In addition, it may prove fruitful to evaluate classroom-level factors using longitudinal designs to test victimization as both an antecedent and a consequence of peer rejection, to evaluate potential reciprocal effects. For example, being victimized in a class that supports bullying may make children more rejected over time. In addition, although anti-bullying attitudes were assessed by self-report, peer victimization, peer rejection, and bullying behavior were all assessed through peer reports. Future studies might benefit from utilizing latent variables that reflect assessments from multiple informants, to help reduce possible problems associated with shared method variance (and measurement error).
The current study was also limited in scope in terms of risk factors for victimization. The effects of classroom norms on victimization need to be examined for other risk factors for victimization such as internalizing, externalizing, and physical weakness (Egan & Perry, 1998; Hodges et al., 1997; Olweus, 1978). Additional contextual factors such as class levels of intolerance toward deviations or pervasiveness of prosocial behavior may further elucidate classroom conditions responsible for the translation of risk into actual victimization.
Conceptualizations such as Bronfenbrenner's (1989) ecological systems theory or a cognitive/developmental-ecological model of violence (Guerra, Eron, Huesmann, Tolan, & Van Acker, 1997; Tolan, Gorman-Smith, & Henry, 2003) may be able to provide a more comprehensive framework for understanding the development of behaviors. According to ecological systems theory, behaviors emerge from multiple contextual factors, including individual factors, family factors, extra-familial or community factors, and the macro context (e.g., cultural and societal beliefs). The current study finds support for the notion that, among girls, pro-bullying norms in the classroom may enhance risk of victimization for rejected child, and support the notion that the classroom is a context that may play a meaningful role in girls' behavior. Future studies may benefit from examining additional contextual layers that may promote or mitigate risk for vulnerable children. For example, community norms for aggressive behavior may wield their influence on children. There is already evidence that children who witness high levels of community violence not only exhibit higher levels of post-traumatic symptoms, but also engage in higher levels of aggressive (McDonald & Richmond, 2008) and escalating antisocial behaviors (Miller, Wasserman, Neugebauer, Gorman-Smith, & Kamboukos, 1999). In addition, good family functioning helps to mitigate some of the ill effects of community violence (Gorman-Smith, Henry, & Tolan, 2004; Miller et al., 1999), suggesting that experiences in both the community and the family help to shape behavior in children. By incorporating this type of research we can more comprehensively evaluate the full array of factors that may contribute to victimization (Card & Schwartz, 2009; Espelage & Swearer, 2009). As victimization has been shown to be a multilevel phenomenon, to deepen our understanding it will be important to have continued work on the multilevel dynamics involved in peer victimization.
Appendix
We present the models in the form of hierarchical linear models, using the notation of Raudenbush and Bryk (2002). The dependent variable for all models is the victimization score Victimizationij for child i in classroom j.
To test the risk-resiliency model with cross-level moderator effects, separate models were specified for two different ways of measuring bullying-related classroom norms, namely by the average level of bullying going on in a classroom and by the classroom mean of anti-bullying attitudes. Bullying was scaled in such a way that a score of 0 corresponds with absence of bullying. Anti-bullying attitudes were centered around the sample mean.
To study differential gender effects we compared a model including gender-specific regressions of victimization on rejection with a model specifying this regression to be the same for boys and girls. In addition, classroom norms were differentiated by distinguishing same-sex and cross-sex classroom norms. These were operationalized by the classroom means of bullying frequency and anti-bullying attitudes for boys and girls separately.
Note that we have two equations for the slopes of victimization on rejection, one for girls and one for boys, but only one equation for the intercepts. We assume that the boy–girl difference in victimization, the β1 parameter of the model, is fixed. This means that we allow a boy–girl difference in the level of victimization; but we do not allow the size of this boy–girl difference to depend on the particular classroom. A preliminary analysis showed that this is a plausible restriction. This restriction makes the model computationally feasible. Without the restriction, the estimation iterations resulted in an inadmissible solution, because of a problem with the data for girls. Note that in this model with restrictions on the intercepts, the intercepts for boys and girls still may differ, but this difference is restricted to be the same for all classrooms; so, the gender difference in victimization is considered to be fixed and not randomly varying across classrooms. The variation in the intercepts β0j does represent the classroom-level variation in level of victimization, controlling for all predictors in the model.
With this model we can test for classroom differences in slopes of victimization on rejection separately for girls and boys, indicated by the variances of the u2j and of the u3j (gender-specific effects of social context). Also, averages of intercepts and slopes were allowed to differ between genders. In addition, the model included at the classroom level the covariances of the girls' and the boys' slopes with the intercepts, and the covariance between the slopes of rejection for girls and boys.
Note that the gender-specific model includes four extra parameters compared with the restricted model, which still allows for average gender differences in level of victimization.
The terms ‘Boyclass’ and ‘Girlclass’ refer to the same-sex class norms with respect to bullying. The same model was estimated with bullying replaced by anti-bullying attitudes. The variables boyclass and girlclass can also be inserted in the equation for the within level but were omitted here, because they were not statistically significant, and their removal did not change the other results.
Table 3 presents, next to the estimated fixed effects, the estimated variance components, one at the within-classroom level, and three at the between-classroom level. The two covariances of slopes with intercepts and the covariance between the slopes for boys and girls were included in the model, but are not reported in Table 3.
To examine the moderator effects of cross-sex norms, the variable boyclass was added as a predictor to the equation for β2j, the girls' slopes of victimization on rejection, and the variable girlclass was added to the equation for the boys' slopes β3j.
To examine cross-level interactions in the restricted model with gender-invariant regression, the same-sex class norms were added as a single predictor to the equations at the classroom level using girlclass for girls' scores and boyclass for boys' scores, thereby restricting the effect of class norms to be the same for boys and girls. Finally, a model was specified in which the cross-level interaction is completely the same for boys and girls, using whole-class norms as a predictor of classroom intercepts and slopes. In this model, reported in Table 2, only a main effect of gender on victimization was included.
To further investigate the estimation problem for the model with randomly varying intercepts for both boys and girls, we fitted the model with same-sex norms separately for boys and girls. For the girls' data, further restrictions were needed to fit this model because of an inadmissible covariance between intercepts and slopes. The two possible solutions were restrict the intercept variance to be equal to zero or restrict the slope variance to be equal to zero. Regardless of the restriction applied, the result was that a statistically significant cross-level interaction of same-sex norms and rejection on victimization was obtained for girls, and not for boys, which points in the same direction as the results obtained in Table 3. Note that the restrictions needed in the separate model for girls are more extreme than fixing the slope for the dummy variable ‘Boy’ in Table 3; nevertheless, the results for the cross-level interaction pointed in the same direction in all cases, which indicates that the conclusions obtained are robust.
