Does Social Withdrawal Inhibit Defending Bullied Peers and Do Perceived Injunctive Norms Mitigate Those Effects?

Abstract

Although a number of social-cognitive and contextual correlates of defending against bullying have been identified, research on the personality traits associated with defending have yielded weak and inconsistent results. The current study provided a novel examination as to whether a tendency toward social withdrawal is associated with less frequent defending and whether perceived injunctive norms for defending and aggression minimize the impact of social withdrawal on defending behaviors. A sample of 1,564 children (760 girls; M_age = 10.05; 55.0% White; 36.1% Black) were followed in the fall, winter, and spring of a school year. Defending was measured with self-reports and peer-reports. Social withdrawal was measured using teacher-reports. Perceived injunctive norms were estimated by calculating within-person correlations between participants’ ratings of peers’ popularity and defending and between peers’ popularity and aggression. Results revealed that social withdrawal was associated with less defending in the fall, and this effect was sustained over the school year. For boys, lower levels of social withdrawal in the fall were associated with less defending when they viewed popular peers as unlikely to defend. A temporary (i.e., fall) association was found between viewing defenders as popular and self-reported defending, and children became less likely to self-report defending over the school year if they viewed popular children as aggressive. These findings underscore the need to examine how temperamental traits and perceived contextual norms co-contribute to bystanders’ behavior when witnessing bullying.

Dating as far back as the work of Dan Olweus (1978), investigators have been acutely aware of the prevalence of bullying and its detrimental consequences for those who are peer victimized. This prompted extensive research into the characteristics of children who aggress against peers or are the victims of bullying. A marked shift in the study of aggression and peer victimization occurred with the recognition that bullying often involves the larger peer group, including witnesses who observe the bullying and often take an active role (see Salmivalli et al., 1996; Sutton & Smith, 1999). Consequently, the past two decades has seen a precipitous increase in studies focused on the bystanders to bullying. Of the bullying roles identified, defending has received particular attention due its potential positive impact on bullying prevention (Salmivalli et al., 2005) and the need, therefore, to understand what differentiates the minority of students who defend (Goossens et al., 2006; Salmivalli et al., 1996) from other bystanders.

Much of the research on defending has elucidated social-cognitive, emotional, and interpersonal factors that underlie defending (e.g., empathy, moral disengagement, social status; Ettekal et al., 2015; Meter & Card, 2015a). A less prominent line of research has focused on how personality traits, particularly the Big Five, correlate with defending (e.g., Mazzone & Camodeca, 2019; Pronk et al., 2015; Tani et al., 2003). However, findings from these studies have been mixed, often yielding weak associations at best. One limitation of studying the Big Five personality traits is that they encompass a range of behaviors and affective reactions likely unrelated to defending when witnessing bullying. The current study, therefore, examines a lower-order personality trait that potentially has direct consequences for defending – social withdrawal (i.e., an asocial disposition reflecting a preference for solitude and low social approach motivation, Rubin & Asendorphf, 1993). Specifically, this investigation had two primary objectives: (a) to examine concurrent and longitudinal associations between social withdrawal and defending and (b) to determine whether perceived injunctive norms moderate these links. We propose that whether being socially withdrawn inhibits defending may depend on the extent to which the broader peer context is seen as supportive of defenders or supportive of the children who engage in aggression. Thus, children’s perceptions of the social status, in this case popularity, of peers who they identify as defending or aggressing is examined as a potential moderator of the association between social withdrawal and defending.

Defending and the Potential Role of Social Withdrawal

Behaviors such as supporting the victim, telling an adult, or making the bullying stop are classified as defending in bullying situations (Salmivalli et al., 1996). Evidence of the potential salutary effects of defending has spurred increased attention to its association with children’s psychosocial development. For example, Salmivalli et al. (2011) found that higher levels of defending within a peer group is associated with lower rates of bullying and risk for peer victimization, and perceiving peers as defending has been shown to predict reduced bullying behavior during the school year (Saarento et al., 2015). Peer-victimized children who report having defenders evidence better psychosocial adjustment than those without defenders (Ma & Chen, 2019; Sainio et al., 2010), consistent with the finding that the association between peer-victimization and depressive symptoms is weaker in classrooms in which confronting aggressors is common (Yun & Juvonen, 2020). Although stronger, particularly longitudinal, evidence is needed to determine whether increasing defending leads to lower rates of bullying and better adjustment for peer-victimized children, fostering defending has become a focus of many anti-bullying interventions (Polanin et al., 2012; Salmivalli et al., 2005). In order to optimize such anti-bullying efforts and the potential benefits of defending, there is a need to identify characteristics of children that promote defending behaviors. To this end, a substantial literature has been amassed focusing on the socio-cognitive (e.g., self-efficacy, moral disengagement) and affective (e.g., empathy) correlates of defending (see Ettekal et al., 2015; Meter & Card, 2015a).

Drawing from the personality psychology literature, researchers have examined whether personality traits, particularly extraversion, are associated with defending, yielding inconsistent results. Pronk et al. (2015) and Mazzone and Camodeca (2019) found no association between extraversion and defending among early adolescents. In contrast, using a person-centered analysis, Tani et al. (2003) found defenders in middle childhood have higher levels of energy, a component of extraversion, than outsiders, and Janošová et al. (2018) found a small positive association between defending and extraversion for early adolescent girls, but not boys.

These weak and inconsistent associations can be attributed to a number of factors. Although a small set of personality traits are evident in childhood and adulthood (Soto & Tackett, 2015), personality is often conceived of as hierarchically organized, subsuming a set of lower order traits (Shiner & Caspi, 2003). Examining lower order traits can provide insights into individual differences lost when studying only higher-order personality factors (Paunonen & Ashton, 2001). For example, lower-order traits are differentially associated with aggression and antisocial behavior (Jones et al., 2011) and various forms of psychopathology (Watson et al., 2019). Furthermore, studies relying solely on the Big Five personality traits fail to take into consideration contextualized views of personality that account for the contingent nature in which personality traits arise (e.g., “my shyness emerges when around new people”; McAdams, 1992).

Thus, more consistent links between extraversion and defending may emerge by examining behavioral dispositions that are narrower in nature and more context-specific than the Big Five. Social withdrawal (i.e., low sociability) within peer contexts may be a particularly potent correlate of defending peers. Defending requires active engagement with the peer group (i.e., an approach-oriented behavior) and a willingness to risk one’s own social well-being for others. For example, defending peers can elicit aggression from peers (Huitsing et al., 2014) and lead to reduced peer acceptance (Meter & Card, 2015b). Unsurprisingly, children report concerns that defending others would elicit negative attention and lead to their own peer victimization and marginalization (Strindberg et al., 2020). There is also evidence that defending is associated with emotional distress (Lambe et al., 2017) and predicts heightened depressive symptoms among low-status, peer-victimized youth (Malamut et al., 2021). Children who are oriented toward greater peer engagement and social mastery goals (Caspi et al., 1988; Shim & Ryan, 2019) may be able to overcome such challenges and act on behalf of a victim. In contrast, children who are oriented “away from the world” (Caspi et al., 1988) withdraw from peer contexts, likely including those that could lead to conflict with others, such as witnessed bullying. Furthermore, because socially withdrawn children are motivated to avoid peer disapproval (Ryan & Shim, 2008), they may determine that the social risks inherent in defending outweigh moral obligations to help others and any potential social rewards that may come with aiding peers.

Despite a strong theoretical basis for expecting a negative association between socially withdrawn behavior with peers and defending, only a small number of studies have tested this association. Perren et al. (2012) found no association between peer-reported defending and teacher-reports of preschoolers’ social participation and a very modest association (r = .10) with parent-reported social participation. Camodeca et al. (2015) found a somewhat stronger positive association between defending and observed social competence, a construct that included skillfully interacting with peers, in a sample of preschool children, and a small positive association emerged between assertiveness and defending in a study of middle schoolers (Jenkins et al., 2016). The strongest evidence comes from a study of 9^th through 12^th graders showing that youth identified as defenders by their peers received fewer peer nominations for social withdrawal than those nominated as outsiders or victims (Pouwels et al., 2016). Thus, although there is some indication in the extant literature that actively engaging with peers is positively related to defending, this research is limited in that primarily preschoolers or adolescents have been studied, and social withdrawal has rarely been specifically assessed.

A second limitation of the extant research on personality traits and defending is the sole examination of concurrent associations. Rank-order differences in defending may be associated with a range of individual difference variables only modestly correlated with personality traits (e.g., friendship ties, empathy; Oldenburg et al., 2018; Pozzoli & Gini, 2010). Social withdrawal, however, may predict within-person changes in defending, as asocial children may become increasingly reluctant to intervene on behalf of peer-victimized classmates. Longitudinal studies are needed to determine whether lower-order traits, such as social withdrawal, predict within-person shifts in defending. The current study addresses these gaps in the literature by examining whether social withdrawal in the fall: (a) is associated with between-person differences in defending in the fall, (b) predicts within-person changes in defending across a school year, and (c) forecasts between-person differences in defending at the end of the school year.

Perceived Injunctive Norms, Social Withdrawal, and Defending

Weak associations between defending and extraversion may also reflect context-dependent associations. It has long been theorized and documented that the expression of personality traits fluctuates as a function of psychological and contextual contingencies (Brown & Moskowitz, 1998; see also Dirks et al., 2012, for a discussion of situational variability of behavior among children). Furthermore, individual differences in perceptions of context can account for situation-dependent variability in behavior (Shoda et al., 2002; Tennen et al., 2005). Particularly salient during childhood and adolescence are group norms, standards for behaviors or attitudes, reflected in both the base rate at which behaviors or attitudes are displayed (i.e., descriptive norms), as well as perceived attitudes as to what is viewed as acceptable by group members (i.e., injunctive norms; Cialdini et al., 1991). Previous research has shown considerable between-classroom differences in levels of bullying, peer victimization, and defending and, more importantly, significant associations between classroom perceived injunctive norms and members’ bullying role behaviors (Pozzoli et al., 2012; Troop-Gordon et al., 2019). Self-reported perceived injunctive norms are also associated with children’s bystander behaviors. For example, children who report that their peers approve aggression and victimization evidence low levels of defending behavior (Lucas-Molina et al., 2018). In addition, children who view their peers as supporting intervention on behalf of the victim engage in low levels of passive bystanding behavior (although no association was found with defending; Pozzoli et al., 2012).

Investigators have also examined injunctive norms as evidenced by which behaviors in a peer group are associated with social rewards, particularly peer acceptance or popularity (Dijkstra & Gest, 2015; Peets et al., 2015). Popularity, a form of social status characterized by prominence within the peer group, leadership, and social attractiveness (van den Berg et al., 2020), is a strong motivator as children move from childhood into adolescence (Rodkin et al., 2013). Thus, popular youth are a crucial source of information as to which behaviors are desirable (Brechwald & Prinstein, 2011) and considered acceptable by the peer group. Consequently, the behaviors associated with popularity can impact social interactions and relationships throughout the peer group. For example, in classrooms where popularity is more positively associated with bullying, aggressive behavior is less likely to be associated with social rejection (Dijkstra et al., 2008), and children are less likely to defend peers (Garandeau et al., 2019). In addition, Peets et al. (2015) found that, with the exception of very low popular youth, children are more likely to defend if popularity in their classroom is inversely related to bullying.

Perceived injunctive norms may also moderate the link between social withdrawal and defending behavior. Socially withdrawn children are highly sensitive to the potential for rejection and social censure (Wichmann et al., 2004). Therefore, socially withdrawn youth may deem the bullying context as too risky for them to engage in defending (Huitsing et al., 2014; Strinberg et al., 2020). Viewing popular peers as engaging in defending behaviors, rather than aggressive behaviors, may alter this perception of the social context and their assessment of the relative risk of attempting to defend a victim of bullying. Thus, the extent to which socially withdrawn children defend peers may depend on whether they perceive popular classmates as likely to defend peers or to act aggressively towards others.

Previous research has assessed injunctive norms by calculating within-class correlations between social status and the target behavior (Peets et al., 2015). However, the correlation estimated at the group level may not reflect each child’s perceived injunctive norms. For example, not all children agree as to which peers are popular or which act as defenders. A more proximal assessment of perceived injunctive norms would assess the association between the popularity and behaviors the child ascribes to each of their peers (i.e., the within-child correlation). This may be particularly important for socially withdrawn children, who likely base their perceptions of peers on more limited social experiences than less socially withdrawn children, and, therefore, may be less concordant with other children in their perceptions of the peer group. Thus, the final objective of this study was to examine whether concurrent and longitudinal associations between social withdrawal and defending are moderated by children’s perceived injunctive norms for defending and aggression.

The Current Study

Our primary research goal was to examine whether fall social withdrawal is (a) associated with defending behaviors in the fall and (b) predictive of defending trajectories across the school year. To achieve this objective, data were utilized from a study of fourth-grade and fifth-grade children followed across three time points during a single school year. We hypothesized that higher levels of withdrawal would be associated with lower levels of defending in bullying situations cross-sectionally and longitudinally.

The second goal was to examine whether perceived injunctive norms moderate concurrent and longitudinal associations between social withdrawal and defending. We expected that social withdrawal would be more positively associated with defending when children viewed classmates they perceived as popular as being high in defending and low in aggression. Although not a primary goal, analyses also allowed for a third research goal -- testing the main effects of perceived injunctive norms. We anticipated that perceiving popular peers as defenders would be associated with higher levels of defending in the fall and increases in defending throughout the school year. Perceiving popular peers as aggressive was expected to be associated with lower levels of defending in the fall and decreasing levels of defending throughout the school year.

Our fourth goal was to examine the association between social withdrawal in the fall and continued or emerging differences in defending by the end of the school year. It is possible that social withdrawal is associated with lower levels of defending in the fall, but differences dissipate by the spring (e.g., those high in social withdrawal increasingly defend over the course of the school year, becoming as likely to defend as those low in social withdrawal). It is also possible that fall social withdrawal and injunctive norms are not immediately associated with defending, but shape trajectories of defending such that differences emerge by the end of the school year. Thus, our analytic strategy included an exploratory examination of potential transient, stable, or emerging associations between social withdrawal and defending.

Our last goal was to explore potential gender differences in the association between social withdrawal and defending and in the moderating role of perceived injunctive norms. Although previous research has not provided evidence of gender differences in the associations being tested, there are reasons to hypothesize stronger links for boys than for girls. Girls are more likely than boys to defend peers (Goossens et al., 2006) and engage in prosocial behaviors (Kuhnert et al., 2017). Therefore, girls may feel implicit pressure to conform to gender norms and defend peers despite any predisposition towards social withdrawal, weakening the association between social withdrawal and defending. Evidence also suggests that boys are more concerned with popularity and social status than girls (LaFontana & Cillessen, 2010) and are less resistant to peer influence (Steinberg & Monahan, 2007). Thus, perceptions of popular peers may have a greater influence on boys than girls. Accordingly, multi-group path analyses were conducted allowing for an examination of gender differences in all concurrent and predictive associations. Finally, perceptions of popular peers may impact children’s motivation to see themselves as defenders of bullied peers more than their actual defending behaviors. Therefore, we examined associations with self-reported and peer-reported defending to test for the robustness of effects across reporters and to identify potential unique associations.

Method

Participants

Data for this study came from 1,564 children (M_age = 10.05 years; SD = .67; 760 girls) in 4^th and 5^th grades who had attended one of 13 public elementary schools located in small, rural, primarily low income communities in the Southeastern United States. At the beginning of the school year, all children in the schools’ 91 classrooms were invited to participate. Class sizes ranged from 13 to 28. Of these children, 1,564 (76.1%; SD = .13; Min = 34.6%; Max = 100%) received written parental permission and provided written assent to participate (36.1% Black; 4.1% Latina/o; 55% White; 0.7% Asian; 0.1% Native American; 0.2% Native Hawaiian/Pacific Islander; 3.5% Multiethnic; 0.2% Other). The percentage of children at each school receiving reduced or free lunches ranged from 53.1% to 93.6% (M = 71.31%). Classroom teachers were predominantly White (78.9%; Black = 17.8%; Latina/o = 3.3%).

Measures

Social withdrawal.

To measure social withdrawal, teachers completed the six-item Asocial with Peers subscale from the Child Behavior Scale (CBS-AS; Ladd & Profilet, 1996). Participating students were rated by their teachers on a scale from 1 (Not True as Far as You Know), 2 (Somewhat or Sometimes True), to 3 (Very True or Often True). Example items include “likes to be alone” and “avoids peers.” Ladd et al. (2009) provided extensive evidence as to the internal reliability, construct validity, and criterion-related validity of the CBS-AS for Grades 1–8. A composite social withdrawal score was created by averaging the six items (α = .93). Although social withdrawal was assessed at all three waves of data collection, only fall data were used for these analyses to focus on prospective relations with defending.

Defending behavior.

Self-reported and peer-reported defending behavior was measured with the Participant Response Questionnaire (PRQ-SR; Salmivalli et al., 1996; Salmivalli & Voeten, 2004). To obtain self-reports, children responded to five items asking how often they engaged in defending behaviors. These items were derived from the 15-item version of the PRQ measure developed by Salmivalli and Voeten (2004). Two of the items retained the original wording (“tell others to stop bullying,” “try to make others stop bullying”). To increase the measure’s internal reliability, the third item was split into two items (“comfort the victim,” “encourage the victim to tell the teacher about the bullying”), and an item from the original PRQ was revised to capture active efforts to solicit adult intervention (“tell the teacher or another adult about the bullying”). Previous research has established that self-reported defending correlates significantly with peer-reports (e.g., r = .46; Salmivalli et al., 1996). Children responded to each item on a scale from 1 (Never) to 4 (A lot). A composite defending score was computed by averaging the item scores (α = .75, .81, and .84, for fall, winter, and spring, respectively).

To obtain peer-reports of defending, children rated each of their participating classmates on three PRQ items (“tells the others to stop bullying,” “tries to make the others stop bullying,” and “comforts the victim or encourages him/her to tell the teacher about the bullying”) on a scale from 1 (Never) to 4 (A lot). These items have evidence good internal reliability (α = .89) and criterion validity in previous research (Salmivalli & Voeten, 2004). For each item, the average rating received was calculated. A composite defending score was calculated by averaging the three item-level scores (α = .92, .94, and .94, for fall, winter, and spring, respectively).

Perceived defender popularity.

To measure children’s perceptions of the popularity of peers they identify as defenders, a correlation coefficient was computed between the ratings a child gave participating classmates on the defending items and the ratings the child gave classmates on an item measuring popularity. The defending items were the same as those used to determine a child’s peer-reported defending. However, instead of computing the average ratings received, we calculated the average rating a child gave a peer across the three defending items. To assess children’s view of the popularity of each of their classmates, we utilized an item that asked children to rate how popular each of their participating classmates was on a scale from 1 (Not at All) to 3 (A lot). Popularity was defined for the children as being respected by other children, seen as being “cool,” and having many kids want to be friends with you. This definition has been used previously (Troop-Gordon & Ranney, 2014) and is consistent with commonly identified features of popularity (e.g., Rose & Swenson, 2009; Ryan & Shim, 2008). This item has been used in a number of studies, yielding evidence of discriminant and predictive validity (e.g., Troop-Gordon & Ranney, 2014; Troop-Gordon et al., 2019). A within-individual correlation coefficient was calculated for each child by correlating the composite defending rating given to each classmate and the popularity rating given to each classmate.

However, perceived defending popularity correlation coefficients could not be calculated if a child: (a) gave the same composite defender rating to all classmates, yielding no variability on that measure, (b) gave the same popularity rating to all classmates, yielding no variability on that measure, or (c) for either defending or popularity provided ratings for less than half of their class. This led to missing data on this item for 31.52% of the sample.

Perceived aggressor popularity.

A similar procedure was employed to measure children’s perceptions of the popularity of peers they identified as aggressive. A correlation coefficient was computed between the ratings a child gave participating classmates on three peer-report items assessing aggression and the rating the child gave classmates on the popularity item. Peer-reported aggression was measured with items tapping physical, relational, and verbal aggression (“hit or push other kids,” “tell other kids they can’t play with them or that they won’t be friends with them,” “call other kids bad names or say mean things to them”) on a scale from 1 (Never) to 4 (A lot). These items have been used in previous studies (e.g., Rose & Swenson, 2009; Troop-Gordon & Ranney, 2014) and showed good internal reliability in the current study (α = .93). The ratings a child gave each classmate on these three items were averaged, and a within-individual correlation coefficient was calculated by correlating the composite aggression rating given to each classmate and the popularity rating given to each classmate. As with the perceived defender popularity variable, scores could not be provided if there was no variance in the aggression or popularity ratings given or if a child rated less than half of the participating classmates on either item, resulting in missing data for 31.71% of the sample on this item.

Procedures

Data were collected in two consecutive years. The first cohort of five schools participated in the 2017–2018 school year. The second cohort of eight schools participated in the 2018–2019 school year. The two cohorts came from neighboring counties and were geographically and economically similar. Schools included fourth grade and fifth grade classes with the exception of one school that went only to fourth grade. All children participated in a randomized controlled trial of a theory-based anti-bullying program that sought to increase defending behaviors when witnessing bullying, and, in turn, improve children’s emotional well-being, school adjustment, and academic progress. Prior to initiating consent procedures or data collection, all schools were randomly assigned to either the novel intervention based on Deviance Regulation Theory (DRT; Blanton & Burkley, 2008) or a control intervention. The DRT intervention focused on increasing positive perceptions of defenders, while the control condition focused on increasing empathy for victims of bullying. With the exception of the targeted processes, all intervention activities were identical (e.g., a class discussion as to how one can defend peers, a poster-making project). Eight of the schools were assigned to the DRT intervention (55 novel intervention classrooms; 37 control intervention classrooms). In all schools, children participated in a fall data collection and two weeks later in a 45-minute intervention activity. Therefore, the intervention could not have influenced concurrent associations in the fall. The winter data collection took place approximately three months after the intervention activity, and the spring data collection took place three months after that. Controlling for intervention condition did not alter the direction or significance of the findings for any of the longitudinal analyses. Therefore, analyses are presented in which intervention condition was not controlled.

Children completed group-administered questionnaires in their classroom. Data collection took approximately 50–55 minutes. An undergraduate or graduate research assistant read the instructions and the questions aloud, and two or more assistants were present to provide help with reading the items or directions. At each wave of data collection, teachers completed a packet of questionnaires for each participating student. They received a $10 USD honorarium for each packet completed at each wave of data collection. This study was approved by the Institutional Review Board of Auburn University, Protocol #17–092 MR 1703, Project Title, “Using Deviance Regulation to Combat Bullying.”

Transparency and Openness

All deidentified data, measures, and Mplus code are available upon request from Dr. Troop-Gordon. This study was not preregistered.

Data Analytic Strategy

Two sets of analyses were conducted to test associations between social withdrawal and defending, as well as the moderating role of perceived defender popularity and perceived aggressor popularity. In the first set, two multi-group models were estimated to examine concurrent associations between social withdrawal and defending in the fall, the main effects and moderating role of injunctive norms, and potential gender differences in these main and interactive effects. One model tested associations with self-reported defending, and the other tested peer-reported defending. Social withdrawal, perceived defender popularity, and perceived aggressor popularity were included simultaneously as predictors, as was a social withdrawal × perceived defender popularity interaction and a social withdrawal × perceived aggressor popularity interaction. All variables were grand-mean centered prior to inclusion in analyses and the creation of interaction terms. To account for the clustering of children in classrooms, multi-level analyses were employed and means of defending were estimated at the classroom level. To identify gender differences, tests for invariance of the paths from each predictor to defending was conducted using the Satorra-Bentler scaled χ² difference test (Satorra & Bentler, 2010). When a significant difference was detected after constraining all parameters to be equal across gender, individual paths were sequentially freed to identify those for which boys and girls differed. When a significant interaction was detected, simple slopes for social withdrawal were plotted at low (−1 SD), average (0 SD), and high (1 SD) levels of the moderator.

A second set of analyses were conducted to test longitudinal associations between social withdrawal and defending trajectories across the school year, the main and moderating effects of injunctive norms in the prediction of defending trajectories, and gender differences in the main and interactive effects. To accomplish these goals, growth trajectories were estimated, separately for self-reported and peer-reported defending, anchoring the intercept at the fall data collection (i.e., the linear growth term was set as 0, 1, and 2, for the fall, winter, and spring, respectively). To account for clustering, multi-level analyses were employed and growth parameters were estimated at the within-classroom and between-classroom levels. All main effects and interaction terms were set as predictors of the within-classroom latent intercept and linear growth term, and gender differences in the within-classroom latent growth parameters and in all predictive associations were tested. When significant associations were found, trajectories were estimated at low (−1 SD) and high (1 SD) levels of the applicable predictors.

To accomplish our goal of examining whether social withdrawal and injunctive norms in the fall were predictive of spring defending, we reran the growth curve analyses with the linear growth terms set at −2, −1, and 0. This allowed us to detect whether significant associations identified in the fall were sustained or whether they dissipated due to differences in the slopes. It also allowed us to identify associations between the predictors and defending that may have gradually emerged during the school year.

Results

Preliminary Analyses

Missing data analysis.

Missing data for social withdrawal, self-reported defending, and peer-reported defending ranged from 1.40% to 9%. There was significantly more missing data on the perceived defender popularity and perceived aggressor popularity variables due to children not providing peer-report data or giving all peers the same rating on popularity, defending, or aggression. To examine potential bias due to missing data, we compared those children with missing data on social withdrawal and/or the defending variables (18.1%) to those with no missing data on gender and all other study variables. Children with missing data had higher social withdrawal scores than those with no missing data [t(1,483) = 3.82, p < .001; M_Missing = 1.23, SD_Missing = .44; M_NoMissing = 1.13, SD_NoMissing = .44] and lower levels of peer-reported defending [Wave 1 = t(1,540) = −3.12, p < .01; M_Missing = 2.41, SD_Missing = 0.51; M_NoMissing = 2.51, SD_NoMissing = 0.48; Wave 2 = t(1,508) = −4.95, p < .001; M_Missing = 2.37, SD_Missing = 0.52; M_NoMissing = 2.55, SD_NoMissing = 0.49; Wave 3 = t(1,483) = −3.38, p = .001; M_Missing = 2.41, SD_Missing = 0.55; M_NoMissing = 2.54, SD_NoMissing = 0.51].

We then compared those children with missing data on either the perceived defender popularity or perceived aggressor popularity (33.1%) to those children not missing either of these two variables. Children with missing data were more likely to be boys (61.1%) than girls, 38.9%; χ²(1) = 29.18, p < .001. Compared to children with no missing data, children with missing data had higher social withdrawal scores [t(1,483) = 2.53, p = .01; M_Missing = 1.18, SD_Missing = 0.38; M_NoMissing = 1.13, SD_NoMissing = .33] and had lower self-reported defending in Wave 1, [t(1,504) = −3.22, p = .001; M_Missing = 3.17, SD_Missing = 0.78; M_NoMissing = 3.29, SD_NoMissing = .64] and peer-reported defending at Waves 1 and 2 [Wave 1 = t(1,540) = −2.80, p < .01; M_Missing = 2.44, SD_Missing = 0.48; M_NoMissing = 2.52, SD_NoMissing = 0.48; Wave 2 = t(1,508) = −2.28, p < .05; M_Missing = 2.48, SD_Missing = 0.49; M_NoMissing = 2.54, SD_NoMissing = 0.50]. By using full information maximum likelihood in Mplus (Muthén & Muthén, 1998–2017), all available data were included in each analysis, resulting in ns ranging from 1,556 to 1,564 across analyses.

Descriptive statistics.

Descriptive statistics are presented in Table 1. On average, children reported moderately high levels of defending for themselves and their peers, a trend that was reflected in classroom means for defending. Social withdrawal scores were, on average, low. However, as Wave 2 and Wave 3 data were collected following the DRT and control interventions, means on these variables should be interpreted in light of potential intervention effects. Independent t-tests indicated that boys were more socially withdrawn than girls [M_boys = 1.18, SD_boys = .39; M_girls = 1.11, SD_girls = .30; t(1,483) = 4.16, p < .001], and girls evidenced greater self-reported and peer-reported defending at all three time points (all ps < .001). There were no gender differences in perceptions of defender popularity and aggressor popularity.

Table 1.

Descriptive Statistics

	Boys				Girls
Variable	M	SD	Min.	Max.	M	SD	Min.	Max.
Defending – fall self-report	3.14	.74	1.00	4.00	3.38	.61	1.00	4.00
Defending – winter self-report	3.02	.79	1.00	4.00	3.27	.69	1.00	4.00
Defending – spring self-report	2.91	.83	1.00	4.00	3.23	.71	1.00	4.00
Defending – fall peer-report	2.37	.46	1.09	3.56	2.63	.47	1.24	3.77
Defending – winter peer-report	2.38	.48	1.17	3.84	2.67	.48	1.40	3.80
Defending – spring peer-report	2.40	.50	1.08	3.93	2.66	.51	1.23	3.18
Social withdrawal	1.18	.39	1.00	3.00	1.11	.30	1.00	3.00
Perceived defender popularity	.23	.42	−1.00	1.00	.25	.41	−1.00	1.00
Perceived aggressor popularity	−.10	.44	−1.00	1.00	−.10	.45	−1.00	1.00

For both perceived popular peers’ defending and perceived popular peers’ aggression, mean scores were close to 0, and ranged from −1.00 to 1.00, suggesting substantial variability across children in the extent to which they perceived popular peers as defenders or aggressors. Multi-level unconditional models were also conducted to determine the extent to which variance in defending could be attributed to individual differences at the child or classroom level. Significant within-class and between-class variance was found for both self-reported and peer-reported defending, all ps < .001, but intra-class correlation coefficients (ICCs) indicated that more variance could be accounted for by classroom-level differences for peer-reported defending than for self-reported defending (ICC = .41 and .04, for peer-report and self-report, respectively).

Bivariate correlations.

Bivariate correlations are presented in Table 2. Over the school year, the stability of self-reported defending was moderate, and the stability of peer-reported defending was high. Agreement between self-reports and peer-reports of defending was significant but modest. Teacher-reported withdrawal was not significantly associated with most other variables, evidencing only a small significant negative correlation with peer-reported defending for boys in the spring and with self-reported defending for girls in the fall and winter. Perceived defender popularity and perceived aggressor popularity were correlated with defending in expected directions, although the magnitude of these correlations was often small, and were moderately and negatively correlated with each other.

Table 2.

Bivariate Correlations

Variable	1	2	3	4	5	6	7	8	9
1. Defending – fall self-report	---	.48***	.44***	.15***	.15***	.19***	−.07*	.04***	−.12**
2. Defending – winter self-report	.51***	---	.64***	.22***	.26***	.27***	−.09*	.15***	−.23***
3. Defending – spring self-report	.42***	.60***	---	.23***	.30***	.30***	−.02	.13**	−.19***
4. Defending – fall peer-report	.20***	.21***	.19***	---	.73***	.66***	−.07	.05	−.08
5. Defending – winter peer-report	.23***	.24***	.28***	.72***	---	.86***	−.04	.08	−.11**
6. Defending – spring peer-report	.17***	.23***	.23***	.64***	.86***	---	−.04	.18**	−.15***
7. Social withdrawal	−.01	−.04	−.01	−.04	−.07	−.09*	---	.00	.02
8. Perceived defender popularity	.11*	.11*	.12**	.09*	.03*	.05	−.04	---	−.64***
9. Perceived aggressor popularity	−.04	−.12**	−.11*	−.03	−.04	−.05	.06	−.50***	---

Note. Correlations below the diagonal are for boys. Correlations above the diagonal are for girls.

^*p < .05.

^**p < .01.

^***p < .001.

Fall Defending as a Function of Social Withdrawal, Injunctive Norms, and Gender

Multi-group multilevel models examining associations with fall self-reported defending revealed a significant decrement in model fit when all means, intercepts, and regression coefficients were constrained to be equal for boys and girls, Δχ²(13) = 46.28, p < .001. Releasing equality constraints sequentially revealed two significant gender differences. The mean of fall social withdrawal was higher for boys than for girls [M_girls = −.04, M_boys = .04; Δχ²(1) = 10.14, p = .001], and the intercept of fall self-reported defending was higher for girls than for boys [M_girls = 3.38, M_boys = 3.14; Δχ²(1) = 28.71, p < .001]. Thus, a final model was estimated constraining all parameters to be equal for boys and girls except the mean/intercept of fall social withdrawal and defending. This model fit the data well, χ²(27) = 39.10, p = .06. CFI = .96, RMSEA = .024, SRMR = .068. Regression coefficients can be found in Table 3. A significant main effect of perceived defender popularity emerged. Viewing popular peers as defenders was associated with higher levels of defending. No other main effects or interactions were statistically significant.

Table 3.

Model Predicting Fall Defending from Social Withdrawal, Perceived Popularity Status of Defenders, and Perceived Popularity Status of Aggressors

	Self-report defending		Peer-report defending
	All children		Boys		Girls
Predictor	b	SE	b	SE	b	SE
Social withdrawal	−.08	.06	−.10***	.03	−.10***	.03
Perceived defender popularity	.17*	.07	.02	.03	.02	.03
Perceived aggressor popularity	−.04	.06	.07*	.03	.07*	.03
Social withdrawal × perceived defender popularity	.07	.26	.23**	.08	.04	.18
Social withdrawal × perceived aggressor popularity	.08	.24	.07	.11	.26	.20

^*p < .05.

^**p < .01.

^***p < .001.

Multi-group multilevel models examining associations with fall peer-reported defending revealed a significant decrement in model fit when all means, intercepts, and regression coefficients, were constrained to be equal for boys and girls, Δχ²(13) = 40.14, p < .001. Releasing equality constraints sequentially revealed four significant gender differences. The mean of fall social withdrawal was higher for girls than for boys, [M_girls = .04, M_boys = −.04; Δχ²(1) = 10.32, p = .001], as was the intercept of fall peer-reported defending, [M_girls = 2.65, M_boys = 2.38; Δχ²(1) = 27.60, p < .001]. In addition, gender differences were found for the social withdrawal × perceived defender popularity interaction, Δχ²(1) = 5.12, p = .02, and the social withdrawal × perceived aggressor popularity interaction, Δχ²(1) = 4.38, p = .04. Thus, a final model was estimated constraining all parameters to be equal for boys and girls except those for which there was a significant gender difference. This model fit the data well, χ²(25) = 34.22, p = .10. CFI = .97, RMSEA = .022, SRMR = .065. Regression coefficients can be found in Table 3.

A main effect of perceived aggressor popularity emerged, indicating that children were more likely to be reported as defenders if they perceived popular peers as aggressive. Despite a significant gender difference, the social withdrawal × perceived aggressor popularity interaction was not significant for boys or girls. In addition, a main effect of social withdrawal emerged, and for boys, this effect was moderated by a social withdrawal × perceived defender popularity interaction. For boys, social withdrawal was negatively associated with peer-reported defending at low (b = −.19, p < .001) and average (b = −.10, p < .001), but not high (b = 0.00, p = .96) levels of perceived defender popularity (see Figure 1). As anticipated, at high levels of social withdrawal, boys were reported to defend peers less frequently at low levels of perceiving defenders as popular. Unexpectedly, the highest levels of peer-reported defending were estimated at low levels of social withdrawal and low levels of perceiving defenders as popular.

Figure 1.

Social withdrawal × perceived defender popularity interaction predicting fall peer-reported defending for boys.

Defending Trajectories as a Function of Social Withdrawal, Injunctive Norms, and Gender

Multi-group multilevel models predicting defending growth curves were estimated next. For self-reported defending, a significant decrement in model fit was found when all parameters were constrained to be equal for boys and girls, Δχ²(14) = 48.04, p < .001. Freeing parameters sequentially revealed a significant gender difference in the latent intercept, Δχ²(1) = 43.10, p < .001, and latent slope, Δχ²(1) = 13.45, p < 001. A final model was estimated setting all parameters to be equal for boys and girls except the latent intercept and slope. This final model fit the data well, χ²(42) = 62.46, p = .02. CFI = .98, RMSEA = .025, SRMR_Within = .057; SRMR_Between = .084. The latent intercept for boys (M = 3.14) was lower than that for girls (M = 3.37), and boys evidenced greater declines in defending (M = −.12, p < .001) than girls (M = −.07, p < .001). Final parameter estimates can be found in Table 4.

Table 4.

Model Predicting Defending Trajectories from Social Withdrawal, Perceived Popularity Status of Defenders, and Perceived Popularity Status of Aggressors

	Self-report defending		Peer-report defending
	All children		Boys		Girls
Predictor	b	SE	b	SE	b	SE
Intercept
Social withdrawal	−.10	.06	−.09***	.03	−.09***	.03
Perceived defender popularity	.14*	.07	.01	.03	.01	.03
Perceived aggressor popularity	−.07	.06	.06*	.03	.06*	.03
Social withdrawal × perceived defender popularity	.06	.24	.26*	.09	.00	.18
Social withdrawal × perceived aggressor popularity	.07	.25	.06	.10	.24	.20
Slope
Social withdrawal	.02	.03	.00	.01	.00	.01
Perceived defender popularity	−.01	.04	−.01	.01	−.01	.01
Perceived aggressor popularity	−.07	.04	−.01	.01	−.01	.01
Social withdrawal × perceived defender popularity	.07	.14	−.07	.04	−.07	.04
Social withdrawal × perceived aggressor popularity	.03	.12	−.07	.05	−.07	.05

^*p < .05.

^**p < .01.

^***p < .001.

Findings paralleled those obtained when associations were examined cross-sectionally. Perceiving defenders as popular was positively associated with self-reported defending in the fall. In addition, a negative association (p = .09) was found between teacher-reported social withdrawal and self-reported defending. To assess whether associations were sustained throughout the school year, and whether new associations emerged, the trajectories were re-parameterized such that the intercept was centered (i.e., set to 0) in the spring. Due to subtle differences in the slopes, neither the main effect of perceiving defenders as popular (b = .12, p = .13) or social withdrawal (b = −.05, p = .40) were significant in the spring. However, an effect emerged for viewing aggressive peers as popular (b = −.20, p = .006), which can be attributed to a trend effect of viewing aggressive peers as popular on the linear slope of defending (p = .09). Figure 2a shows estimated trajectories at high (1 SD) and low (−1 SD) levels of perceiving popular children as aggressive. These trajectories were plotted separately for boys and girls to take into account gender differences in the intercept and slope. At high levels of viewing popular peers as aggressive, children evidenced slightly greater declines in their self-reported defending over the course of the school year than they did at low levels of viewing popular peers as aggressive, resulting in significantly lower levels of self-reported defending by the spring.

Figure 2.

Estimated trajectories of: (a) self-reported defending for boys and girls as a function of perceived aggressor popularity and (b) peer-reported defending for boys as a function of the social withdrawal × perceived defender popularity interaction.

For peer-reported defending, a significant decrement in model fit was found when all parameters were constrained to be equal for boys and girls, Δχ²(14) = 35.42, p = .001. Freeing parameters sequentially revealed a significant gender difference in the latent intercept, Δχ²(1) = 41.70, p < .001. There were additional gender differences in the prediction of the intercept by the social withdrawal × perceived defender popularity interaction, Δχ²(1) = 7.63, p = .006, and the social withdrawal × perceived aggressor popularity interaction, Δχ²(1) = 7.38, p = .007. A final model was estimated setting all parameters to be equal except those for which a significant gender difference emerged. This final model fit the data well, χ²(41) = 61.79, p = .02, CFI = .99, RMSEA = .025, SRMR_Within = .051; SRMR_Between = .047. The latent intercept for boys (M = 2.37) was lower than for girls (M = 2.66). Final parameter estimates can be found in Table 4.

Although there was a significant gender difference for the social withdrawal × perceived aggressor popularity interaction on the intercept, this interaction was not statistically significant for girls or boys. As was the case when estimating the fall concurrent model, perceived aggressor popularity was associated with greater peer-reported defending in the fall for boys and girls. However, re-parameterizing the intercept to the spring revealed that, due to a subtle, albeit non-significant negative effect of perceiving popular children as aggressive on the slope, the main effect of perceiving popular peers as aggressive was no longer significant in the spring (b = .04 p = .23). In addition, social withdrawal was negatively associated with peer-reported defending in the fall for boys and girls, and re-parameterization revealed that this effect was still significant in the spring (b = −.09, p =.005). For boys, the main effect of social withdrawal in the fall was qualified by a significant social withdrawal × perceived defender popularity interaction. Trajectories of peer-reported defending at high and low levels of social withdrawal and high and low levels of perceived defender popularity were estimated for boys (see Figure 2b). As was found for the cross-sectional analysis, in the fall, low levels of social withdrawal were associated with more frequent defending at lower levels of perceived defender popularity. In contrast, high levels of social withdrawal were associated with more frequent defending at higher levels of perceived defender popularity. However, due to subtle differences in the slopes (the social withdrawal × perceived defender popularity interaction was marginally significant at p = .10), the social withdrawal × perceived defender popularity interaction was not significant in the prediction of spring defending (i.e., by re-parameterizing the intercept to the spring; b = .13, p = .27). However, the main effect of social withdrawal was retained in the spring for boys and girls, such that peer-reported defending remained higher at lower levels of social withdrawal.

Discussion

Although a clear picture has emerged as to the social-cognitive and affective bases for defending (Ettekal et al., 2015; Meter & Card, 2015a), the literature on the personality correlates has yielded less consistent results (e.g., Pronk et al., 2015; Tani et al., 2003). The current study addressed substantial gaps in the literature that may have been contributing to these conflicting findings by examining: (a) a lower-order personality trait, social withdrawal, theorized to be particularly relevant to defending, (b) longitudinal associations, and (c) perceived contextual norms that may moderate whether social withdrawal hinders defending. The current study was the first to document that children who are more socially withdrawn evidence not only low levels of peer-reported defending in the fall, but also continue to evidence low levels of peer-reported defending throughout the school year. This study also took an innovative approach to measuring perceived injunctive norms by assessing within-child correlations between their perceptions of peers’ popularity and their perceptions of their peers’ engagement in defending or aggressive behaviors. As anticipated, social withdrawal was more strongly associated with lower levels of peer-reported defending when children viewed popular peers as not defending others, although this effect was significant for boys only. The importance of perceived injunctive norms was further underscored by their main effects on self-reported defending. Consistent with the proposition that children model popular peers (Rodkin et al., 2013), children engaged in more defending if they associated popularity with defending or low levels of aggression. These findings indicate that personality differences and perceived norms may serve as hindrances to defending, but also potential targets of anti-bullying interventions.

Concurrent and Longitudinal Associations between Social Withdrawal and Defending

The primary objective of this study was to test concurrent and longitudinal associations between social withdrawal and defending. Consistent with predictions, social withdrawal was associated with lower levels of peer-reported defending in the fall and remained predictive of lower levels of defending into the spring. There are a number of reasons why socially withdrawn children may be less likely to intervene or aid a bullied peer. Perhaps most notably is a lack of assertiveness characteristic of socially withdrawn children (Wichmann et al., 2004) that is requisite for defending during or after witnessing bullying. It is also possible that due to their asocial disposition, and greater likelihood of being socially excluded (Booth-LaForce & Oxford, 2008), withdrawn children are less likely than others to be present when peers are bullied. As social withdrawal tends to be fairly stable (Rubin et al., 2009), it is not surprising that social withdrawal in the fall would continue to be associated with low levels of defending in the spring.

Two additional goals of this study were to determine whether associations between social withdrawal and defending are moderated by perceived injunctive norms or children’s gender. By testing these moderating effects, the findings underscored the importance of identifying contextual contingencies that amplify or dampen associations between behavioral dispositions and defending (Brown & Moskowitz, 1998; Dirks et al., 2012). Specifically, for boys, in the fall, the negative association between social withdrawal and defending was reduced if they viewed popular peers as defenders. Children often follow the lead of more popular peers (Lease et al., 2020). Boys in particular are susceptible to peer influence (Steinberg & Monahan, 2007) and are concerned about their social status (LaFontana & Cillessen, 2010). Furthermore, boys socialize in larger groups (Rose & Rudolph, 2006) and engage in more overt aggression than girls (Card et al., 2008). Socially withdrawn boys, therefore, may have greater opportunities to observe their popular classmates’ defending of peers than girls.

However, this moderating effect dissipated over the course of the school year, resulting in increased defending behavior. This finding should be examined in light of person-oriented analyses showing that children and adolescents are frequently nominated by peers as falling into multiple bullying roles. Defenders are often identified by their peers as also encompassing outsider or victim roles, and boys more than girls are found to be nominated as holding multiple bystander roles (e.g., Chen et al., 2020; Pouwels et al., 2016). Thus, children’s bystander behaviors likely vacillate across contexts and time (Chen et al., 2020). The findings presented here point to the need to study gradual changes that occur in children’s bystander behaviors as they learn to navigate evolving peer groups. For example, as socially withdrawn boys begin a new school year, they may be sensitive to social cues reflecting interpersonal threat or support (Horwath et al., 2013; Tang et al., 2021). If they do not perceive popular children as defending, they may engage in behaviors consistent with an outsider role, such as not taking sides in bullying situations or walking away. Alternatively, as social withdrawal is a risk factor for peer victimization (Pouwels et al., 2016), socially withdrawn boys may be the target of bullying at the beginning of the school year, leading to feeling undefended by higher status peers and unable to defend others. Over the school year, shifts in perceived norms and relationships (e.g., opportunities to witness popular children defending; the development of new friendships) may lead these boys to increasingly defend. Such postulations should be tested in subsequent research that addresses changes in multiple bystander roles. However, despite evidence of some within-person increases in defending, for boys and girls, social withdrawal was associated with lower levels of defending in the fall and spring. Thus, high levels of social withdrawal also may be characteristic of children who occupy only outsider or victim roles.

An important caveat for these findings is that significant associations with social withdrawal were found for peer-reported, but not self-reported, defending. This may be partly due to similarities in measurement (i.e., teacher-report and peer-report, respectively). Greater inter-rater agreement is often found when assessments are made through third-party observers than with self-reports (e.g., Henry & The Metropolitan Area Child Study Research Group, 2006; Younger et al., 2000). Peers may be witnessing the same socially reticent behaviors (i.e., social withdrawal, passivity when witnessing bullying) that teachers see. Children, in contrast, may have a self-serving bias when contemplating their own behaviors (Camerini & Schulz, 2018) and, therefore, overestimate the extent to which they defend peers. Indeed, peer-reported and self-reported defending were only modestly correlated in the current study. Socially withdrawn children may also defend in less public contexts (e.g., when with a small group of friends) or in ways that require less self-efficacy to perform and are less observable to a wide audience (e.g., comforting the victim after the aggression has ended; Clark & Bussey, 2020). As children are more likely to nominate as defenders peers who engage in overt and direct forms of defending rather than peers who perform passive forms of defending (Pronk et al., 2013), defending by socially withdrawn children may go unnoticed by the majority of the peer group.

Main Effects of Perceived Injunctive Norms on Defending

This study also provided for testing the main effects of perceived injunctive norms on defending trajectories. For peer-reported defending, there was a temporary positive association between viewing popular children as aggressive in the fall and concurrent defending. Although opposite of what we predicted, this may reflect observations of how peers respond to popular-aggressive classmates, rather than a reflection of using popular peers as a source of perceived norms. Aggression among popular youth is well documented (Troop-Gordon & Ranney, 2014), and a key feature of popularity is being a highly visible leader within the peer group (Lansu et al., 2014). It is possible that children are highly aware of times when peers stand-up to classmates they perceive as popular-aggressive and are more willing to engage in defending of peer-victimized agemates if they see others standing up to aggressive, high status peers.

There were also two main effects in the prediction of self-reported defending. In the fall only, children reported greater defending if the classmates they viewed as popular were also ones they viewed as defenders. Although this finding is consistent with literatures on perceived injunctive norms (Cialdini et al., 1991), that this association held only for self-reported defending suggests that the effect may have had less to do with actual behavioral engagement in defending and more to do with children’s understanding of what they believe they should do when witnessing bullying. Consistent with this explanation, recent evidence suggests that older children overestimate their own helpfulness (Levy et al., 2021).

However, the extent to which children viewed popular peers as defenders in the fall was unrelated to how often they reported defending their peers in the spring. Rather, how much children in the fall viewed popular classmates as aggressive was predictive of lower levels of spring defending, suggesting a shift in which traits associated with popularity are predictive of children’s self-views as defenders. It is possible that over the course of the school year children recognize that popular peers are unlikely to become less aggressive, as popularity predicts heightened aggression over time (LaFontona & Cillessen, 2002). Thus, the aggressive behaviors of perceived popular classmates may be salient indicators as to the peer group’s injunctive norms for aggression versus defending. Furthermore, children who initially view themselves as high in defending may forego this perception if they develop worsening views of their peer-victimized classmates, who may be unskilled socially, anxious, or aggressive (Reijintes et al., 2010, 2011).

Limitations, Future Directions, and Implications for Practice

The present study utilized a multi-informant, three-wave longitudinal design from a large, ethnically diverse sample of children. Moreover, analyses were conducted that allowed for testing not only concurrent associations but also gradual changes in defending that may result in significant differences over the course of the school year. However, there were limitations. Data on social withdrawal were collected only from teachers. An advantage of using teacher-reported social withdrawal is the increased ability to employ a multi-informant design (e.g., teachers, peers, and self). Social withdrawal is also an observable behavior teachers see within the classroom. However, teacher-reports of social withdrawal correlate only modestly with other informants (e.g., peers, self; Younger et al., 2000). Thus, studies that assess social withdrawal using multiple reporters would strengthen the ability to test the reliability of the findings presented here. Furthermore, the measure of social withdrawal focused on asocial solitary behaviors. Stronger links may have been found if a measure of social withdrawal had been included that assessed shyness or anxious-solitude (Rubin et al., 2009). In addition, there is increasing research on subtypes of defending behaviors (Lambe & Craig, 2020). It is possible that social withdrawal and injunctive norms are more strongly associated with direct forms of defending (e.g., confronting aggressors) than those that are indirect and, therefore, less risky or observable (e.g., comforting the victim; telling the teacher anonymously).

An innovative approach was used to create variables for injunctive norms by measuring the within-person correlation between children’s ratings of their peers’ popularity and their ratings of peers’ defending and aggression. This differed from assessing injunctive norms at the classroom level by correlating nominations/ratings received aggregated across classmates (Dijkstra & Gest, 2015; Peets et al., 2015) and provides a more direct assessment of how an individual assesses the classroom reward structure for particular behaviors. This methodology was limited, however, in that approximately 30% of the sample were missing data on injunctive norms due to having no variance in their ratings of peers’ popularity, defending behaviors, or aggression. This was particularly true for popularity, which was measured with only one item. Inclusion of additional items may have prompted greater reflection as to classmates’ popularity (e.g., visibility, coolness), resulting in greater variance in children’s ratings of peers. Variability may also be increased by the inclusion of a “not popular” item to measure popularity (Cillessen & Marks, 2011) or items that ensure variability (e.g., rank the top five).

The developmental period of late childhood was chosen, as this is a time when defending behaviors decrease (Pozzoli et al., 2012). Research with younger and older children would allow for discerning whether these results replicate during different developmental periods. It will be particularly important to examine these associations during early adolescence when bullying increases (Wang et al., 2009), concern for popularity heightens (LaFontana & Cillessen, 2010), and positive associations between popularity and aggressive behavior strengthen (Bukowski et al., 2000). Given these changes, it might be anticipated that social withdrawal would be an even stronger determinant of defending during adolescence and the moderating role of popular peers’ defending behaviors would strengthen. In addition, although the sample was ethnically diverse, it was primarily rural. Therefore, research is needed that examines these associations with children in urban settings and from higher socioeconomic communities. Moreover, although intervention condition did not moderate the associations found here, it is possible that participation in either intervention had effects on bullying role behaviors that amplified, or dampened, the effects of social withdrawal and perceived injunctive norms. Therefore, studies that examine these associations among children not participating in a program designed to foster defending would provide important information about normative processes and potential intervention effects.

These findings also have applied implications. Despite a proliferation of anti-bullying initiatives, such programs have yielded mixed results (Ttofi & Farrington, 2011). One obstacle to increasing defending may be that such behavior is incongruous with a socially withdrawn disposition. Accordingly, the efficacy of intervention programs aimed at increasing defending behaviors may need to be evaluated with an understanding of the intrapersonal factors that can limit children’s ability to assert themselves and an awareness that universal increases in defending behaviors may be difficult to achieve in light of differences in children’s personality. To overcome this obstacle, anti-bullying interventions may need to emphasize varying avenues for defending victims, including behaviors that may be more readily enacted by children who are socially withdrawn (e.g., providing comfort at a later time, writing a note to the teacher). Alternatively, rather than targeting the entire peer group, which may backfire among social withdrawn children, focusing on increasing defending among popular students may be more efficacious and, as highlighted in this study, may indirectly decrease bullying by promoting defending as normative, reducing perceived risks associated with standing up for bullied peers.

Conclusion

The findings of the current study highlight three previously underexamined principles in the literature tying defending to personality variables. First, such research can benefit from studying lower order traits, such as social withdrawal or low sociability. Second, the study of behavioral dispositions may provide insights into within-person changes in defending not evident when only concurrent associations are examined. Finally, the role of personality can best be understood when assessed in conjunction with perceived contextual constraints and supports. Specifically, for peer-reported defending, social withdrawal was identified as a risk factor for low levels of defending and this was particularly true for boys if they viewed popular peers as infrequent defenders. Furthermore, viewing popular children as defenders, or aggressors, had both immediate and year-long associations with children’s self-reported and peer-reported defending behaviors. Given potential reluctance among socially withdrawn children to defend, efficacy of programs to increase defending may best be evaluated in terms of fostering defending among those who easily engage with the peer group, particularly popular group members whose behavior may facilitate defending among less socially assertive peers.