Journal of Youth and Adolescence

, Volume 39, Issue 9, pp 1041–1052

Clueless or Powerful? Identifying Subtypes of Bullies in Adolescence

Authors

    • Behavioural Science InstituteRadboud Universiteit Nijmegen
  • Antonius H. N. Cillessen
    • Behavioural Science InstituteRadboud Universiteit Nijmegen
  • Ron H. J. Scholte
    • Behavioural Science InstituteRadboud Universiteit Nijmegen
Empirical Research

DOI: 10.1007/s10964-009-9478-9

Cite this article as:
Peeters, M., Cillessen, A.H.N. & Scholte, R.H.J. J Youth Adolescence (2010) 39: 1041. doi:10.1007/s10964-009-9478-9

Abstract

This study examined the heterogeneity of bullying among adolescents. It was hypothesized that bullying behavior serves different social functions and, depending on these functions, bullies will differ in their skills, status and social behavior. In a total sample of 806 8th graders, 120 adolescents (52 boys, 68 girls) were identified as bullies based on peer nominations. An additional group of 50 adolescents (25 boys, 25 girls) served as the non-bully comparison group. Cluster analysis revealed three corresponding bully subtypes for boys and girls: a popular-socially intelligent group, a popular moderate group, and an unpopular-less socially intelligent group. Follow-up analyses showed that the clusters differed significantly from each other in physical and verbal aggression, leadership, network centrality, peer rejection, and self-perceptions of bullying. The results confirm the heterogeneous nature of bullies and the complex nature of bullying in the adolescent peer group.

Keywords

BullyingSocial intelligenceMachiavellianismPopularityRelational aggression

Introduction

Lacking empathy or social understanding and being exceptionally aggressive have traditionally been associated with bullying (Pellegrini et al. 1999; Sutton et al. 1999a). At the same time, other results associate bullying with good understanding of social relationships, social power, and covered aggressive assaults directed at the victim (Salmivalli et al. 2000; Sutton et al. 1999a). This duality has lead researchers to the assumption that bullies are not a uniform group. Indeed, previous studies have examined the heterogeneity of bullying. For example, Salmivalli and Nieminen (2002) made a distinction between bullies and bully-victims (students who bully but are also bullied), and found that bully-victims were the most aggressive subgroup of bullies. Another approach to the heterogeneity of bullying is to examine different roles in the bullying process. Salmivalli et al. (1996), identified roles such as ringleader bullies versus assistants, and associated these roles with social status. These studies describe different types of bullies; their results indicate that bullying behavior varies among adolescents.

In spite of these developments, more can be learned about the heterogeneity of bullying. For example, few studies have explored differences in social skills among bullies and their associations with different social functions of bullying. It may also be the case, consistent with the work of Salmivalli et al. (1996) that different bully roles or functions are associated with differences in social status. Furthermore, social-cognitive skills may differ between types of bullies. Bullies may be found along the range of the social competence continuum and vary in social intelligence. A broader picture of the heterogeneous nature of bullying is needed as our understanding of the behavior of bullies affects the design and effectiveness of interventions (Arsenio and Lemerise 2001; Sutton et al. 1999b).

The goal of this study was exactly this: to examine the social functions of bullying and to identify different types of bullies based on patterns of behaviors, status, and social skills. Below, these constructs that may differentiate bullies from each other are discussed. Specifically, differences between bullies are addressed in terms of social intelligence, Machiavellian beliefs, popularity, and aggression.

Do Bullies Differ in Social Intelligence?

The social information processing framework (SIP) has been used in previous research to understand children’s social behavior and adjustment. In a reformulation of this framework, Crick and Dodge (1994) distinguished six steps (p. 76) in a child’s behavioral response to a social stimulus: (1) encoding of cues, (2) interpretation of these cues, (3) clarification of goals, (4) response access or construction, (5) response decision, and (6) behavioral enactment. According to this model, incompetent, maladaptive behaviors, such as persistent aggression, are the result of deficits in one or more of these steps (Sutton et al. 1999a).

Although the model explains a range of social behaviors (Crick and Dodge 1999), it has primarily been applied to the display of aggression. The SIP framework provides an understanding of the social information processing of children’s aggression in general. Even though the model does not aim specifically at bullying, Crick and Dodge (1999) argued that bullying is a type of aggression, and therefore the model would apply to bullying as well.

Crick and Dodge (1994) proposed that chronic aggression is the result of chronic processing styles. These persisting processing styles are called biases or deficits because they are the tendency to respond in a consistent manner to ambiguous social cues. The SIP framework emphasizes that not all aggression is maladaptive or incompetent. Incidental self defense, for example, is not the result of processing deficits, but is an adaptive response to threatening social situations (Crick and Dodge 1999). But the authors argue that persisting aggression, such as bullying, is the result of a deficient information processing style. They acknowledge that some bullies may be proficient at a particular step of the SIP cycle, but they are deficient in at least one of the six SIP steps. They reject the possibility that completely competent social cognitions can precede maladaptive behavior, such as bullying (Arsenio and Lemerise 2001; Crick and Dodge 1999). A study by Coie et al. (1991) verified this idea. They found that aggressive children, such as bullies, attribute hostile intentions to peers and subsequently react with exaggerated aggression. This indicates that aggressive children incorrectly encode or interpret social cues and exhibit behavioral responses that are not normative.

In contrast, Sutton et al. (1999b) introduced a different theoretical framework that challenges the stereotypical view of the bully. According to these authors, bullying is an interaction between two or more participants in a structured group. Some bullies misuse their skills to manipulate the relationships in the group with the intention to change its structure and obtain more power. Sutton et al. argued that the effectiveness of bullying is determined by the bully’s ability to attribute internal mental states to themselves and others, or “theory of mind” (Bartsch and Wellman 1989). This awareness of the thoughts and beliefs of others enables them to predict and anticipate their behavior (Sutton et al. 1999b). Understanding the perspective of others and being able to manipulate their thoughts demonstrates the presence of social intelligence. Therefore, Sutton et al. (1999b) argued that at least some bullies are socially intelligent and have superior theory of mind skills.

Previous studies indicate that the use of relational aggression sometimes demands social intelligence. For example, Björkqvist and colleagues found that the efficient use of relational aggression requires social intelligence. This sophisticated form of aggression demands understanding social relations (Björkqvist et al. 1992, 2000). Putallaz et al. (2007) also suggested that a lack of social intelligence may lead to an insufficient aggressive attack. These interpretations suggest that some bullies may be socially intelligent, depending on the type of aggression that they use to harass their victims.

These two theoretical models imply different assumptions about the causes of bullying. The SIP framework assumes that bullying is a result of social deficits. Crick and Dodge (1999) describe the bully as socially inadequate and define bullying as maladaptive behavior. Sutton and colleagues see bullying as socially undesirable, but argue that bullying may be very efficient, depending on the context in which it occurs (Sutton et al. 2001). Because both theoretical assumptions are supported by empirical results, it seems incorrect to reject one of them. The solution is to assume that some bullies are socially intelligent and victimize for their personal advantage. Others may be less socially intelligent, and may attribute hostile intent to others or are unable or unwilling to choose an appropriate response (Archer 2001; Crick and Dodge 1994). Therefore, we hypothesize that bullies are a heterogeneous group and can be found at both extremes of the social competence continuum.

Do Bullies Differ in Machiavellian Beliefs?

Beliefs and impressions that we have about others are a result of previous social experiences. These experiences not only affect the impressions of specific others but also our general belief in human nature (Sutton and Keogh 2001). Differences in the perception of the human nature have been described as part of the construct of Machiavellianism (Christie and Geis 1970; Sutton and Keogh 2001). Christie and Geis (1970) characterized the “Machiavellian” individual as someone who restrains emotional involvement in interpersonal relations and has little concerns with conventional standards of morality. The Machiavellian easily lies and deceives and effectively controls interactions with others. They also believe that they can easily manipulate others in social and interpersonal situations (Andreou 2004).

Christie and Geis’ (1970) definition of the Machiavellian presents the image of a person lacking empathy and affect, a characteristic that has also been found in bullies. Gini (2006), for example, found that bullying is associated with moral disengagement and a lack of emotional understanding. Bullies are less empathic than their peers (Gini et al. 2007) and more likely justify the use of aggressive behavior without feeling guilty (Gini 2006). Similarly, Björkqvist et al. (2000) found that aggression is associated with a lack of empathic responsiveness. Andreou (2004) found that bullying was associated with a lack of faith in human nature, and specifically for girls with acceptance of manipulation in social interactions.

Sutton and Keogh (2000) suggested that there is an association between Machiavellianism and relational aggression. Several studies support this expectation. Björkqvist et al. (1992) defined indirect aggression as a way to inflict pain on others without being identified as the perpetrator. Spreading rumors and manipulating relationships (i.e., relational aggression) are such indirect means to provoke others that are also typical Machiavellian traits. Furthermore, Sutton and Keogh (2000) found that bullies hold more Machiavellian beliefs than non-bullies. They proposed that different types of Machiavellianism may exist among different groups. Relationally aggressive bullies, for example, may have much stronger Machiavellian beliefs than physically aggressive bullies.

Thus, previous research suggests that some bullies are effective in controlling and manipulating social situations. Prior studies also indicate that bullies accept inappropriate behavior more easily than non-bullies, and may interpret their behavior as efficient and justified. Therefore, we assume that bullies hold more Machiavellian beliefs than non-bullies, and that bullies who are relationally aggressive in particular distinguish themselves from non-bullies on this measure.

Do Bullies Differ in Popularity?

The traditional method of defining peer status distinguishes five sociometric types (popular, rejected, neglected, controversial, and average). This classification system identifies sociometrically popular children as those who are well-liked by their peers, a status type characterized by prosocial traits (Coie et al. 1982). Recently, a second type of popularity has been proposed, perceived popularity. While sociometric popularity is a measure of peer acceptance and represents a pleasant and prosocial behavior style, perceived popularity indicates social power and reputation and represents a more dominant and aggressive behavioral style (de Bruyn and Cillessen 2006; Parkhurst and Hopmeyer 1998; Rodkin et al. 2000). Thus, although both popularity types indicate high status, they reflect different behavioral profiles.

Bullying has traditionally been linked to peer rejection. It has been suggested that aggression and disruptive behavior, like bullying, is related to low acceptance (Coie et al. 1982; Dodge et al. 1982). Indeed, a negative association between bullying and peer acceptance has been established (Pellegrini et al. 1999). However, recent studies have shown that aggressive children, such as bullies, can be disliked by their peers but also perceived as popular (Caravita et al. 2008; Cillessen and Rose 2005; Farmer et al. 2003). These studies indicate that there are differences among bullies in peer status.

Estell et al. (2007) hypothesized that bullies would be rated as popular but would not be well-liked by their peers. Indeed, they found that bullies were rejected by peers (disliked) or socially controversial (liked by some and disliked by others). At the same time, they were perceived as central members of the peer group and socially well-integrated. Findings by Vaillancourt et al. (2003) also supported the perception of the bully as either a rejected member of the peer group, or a popular member. Therefore, we assume that bullies differ in their popularity.

Do Bullies Differ in Relational Aggression?

Spreading rumors, telling lies, gossiping and social exclusion are sophisticated forms of aggression that are defined as indirect or relational aggression (Björkqvist et al. 1992; Salmivalli et al. 2000). Compared to verbal and physical aggression, relational aggression is less outspoken. It is more difficult to identify the perpetrator, and avoiding accusations of deliberately harming others is exactly the intention of the aggressor (Björkqvist et al. 1992; Garandeau and Cillessen 2006).

It has often been suggested that the type and amount of aggressive behavior differs between genders. Boys are perceived by peers as more overtly aggressive, while girls seem to use more indirect means to harm others (Salmivalli et al. 2000). In contrast, Putallaz et al. (2007) indicated that gender differences in physical and relational aggression are often minimal. That is, girls as well as boys display comparable amounts of relational aggression. However, boys tend to use equal amounts of physical and relational aggression, while girls are more inclined to harm others by indirect means.

Traditionally, social intelligence or social competence is linked to adjustment and aggression to maladjustment (Crick and Dodge 1994). However, Björkqvist et al. (2000) indicated that social intelligence is a neutral concept because it does not exclude the use of aggression. In a conflict situation, a socially intelligent adolescent still has the option to solve the problem aggressively rather than peacefully. Thus, aggression and social intelligence are not mutually exclusive, and the likelihood that they go together is larger when a relationally aggressive solution is chosen. Björkqvist et al. (1992) argued that indirect aggression requires influence in a social network. Indeed, it has been found that although relational aggression is negatively associated with social preference, it is positively associated with perceived popularity (e.g., Cillessen and Mayeux 2004).

Thus, it has been argued that relational aggression requires social intelligence, social power, and the skills to manipulate a group. A good understanding of relationships and a central position in the peer group appear necessary to relationally victimize others (Björkqvist et al. 1992; Sutton et al. 1999b). Xie et al. (2002) found that the use of relational aggression did not lead children to be excluded from the peer group. Instead, high network centrality enabled them to use relational aggression even more. Therefore, it was hypothesized that some bullies are particularly relationally aggressive, especially when they are also central or powerful in the peer group.

The Current Study

The overall goal of this study was to identify types of bullies that differ in social intelligence, Machiavellian beliefs, relational aggression, and popularity. These constructs were selected based on their associations with bullying in recent studies (e.g., Andreou 2004; Sutton and Keogh 2000) and their importance in recent debates on the nature of bullying (see Arsenio and Lemerise 2001; Crick and Dodge 1999; Sutton et al. 1999c, 2001). It was hypothesized that the social function of bullying in the peer group depends on the status and position of the bully in the peer network. Different provisions of bullying behavior will demand different skills and influence levels. Thus, our goal was to identify different types of bullies and the unique behaviors, status positions, and social-cognitive skills characteristic for each.

Method

Participants

Participants were 806 adolescents (353 boys, 453 girls; M age = 13.37 years, SD = .57) in Grade 8 of two secondary schools in The Netherlands. Students were in three secondary school tracks: pre-vocational (159 boys, 187 girls), general secondary (83 boys, 125 girls), and college preparatory (111 boys, 141 girls). The ethnic composition of the sample was 92.9% Dutch/Caucasian, .1% Moroccan, .4% Turkish, and 3.5% of other ethnic or national origin (3.1% did not specify their origin). Grade 8 was chosen because this is an age group in which reports of bullying are particularly high (see Nansel et al. 2001; Olweus 1994). Passive parental permission was obtained through a letter. Parents were informed about the purpose of the study and the voluntary nature of participation. Parents could react to this letter if they did not agree that their child would participate in the study. Students completed a questionnaire in their classroom under the guidance of a researcher who explained the procedures and the confidentiality of the answers.

Peer Nomination Measures

A standard sociometric instrument was used that included nine peer nomination questions. The reference group was the classroom. For each question, students could name an unlimited number of peers from their own classroom. Both same-sex and other-sex nominations were allowed. Self nominations were not allowed.

The following constructs were assessed: perceived popularity (“classmates who are most popular”), rejection (“classmates you like the least”, a continuous measure of rejection), bullying (“is a bully”), physical aggression (“hit, kick or push others”), verbal aggression (“call others names, laugh at them, or threaten to hurt them”), relational aggression (two items: “gossip or speak badly about others,” and “ignore or exclude others”), and leadership (“classmates who take the lead”). For each question, the number of nominations received was counted for each student and standardized to z-scores within classrooms to control for differences in classroom size. For bullying, nominations received were also standardized within gender (see Solberg and Olweus 2003). For relational aggression, the average of the two standardized scores (r = .70) was taken.

A within-classroom friendship nomination was also used. The matrix of best friend nominations in the classroom was analyzed with UCINET 6 (Borgatti et al. 2002) to compute a network centrality score for each student. The Bonacich (1987) power index was chosen as the measure of network centrality. According to this index, students received a network centrality score that is a weighted function of reciprocal friends they have in the classroom, and the power of the peers that they have ties with. The Bonacich power index is generally recognized as an important measure of social network centrality.

Self-Report Measures

Bullying

Students completed the bullying subscale of the Dutch version of the Olweus (1989) Bully/Victim Questionnaire (Sentse et al. 2007). The scale includes five items (e.g., “How often have you bullied others in the last year?”) that are rated on a 5-point scale, ranging from “never” to “several times”. The internal reliability of this scale was good (α = .70).

Machiavellianism

Machiavellianism was measured with a Dutch translation of the 20-item Kiddie Mach. The Kiddie Mach was originally designed by Christie and Geis (1970) to assess Machiavellianism in children and adolescents. A Greek translation of the original measure was recently used successfully by Andreou (2004). The items of this scale measure Machiavellian beliefs (e.g., “Anyone who completely trusts anyone else is asking for trouble”) and are rated on a 5-point scale (1 = disagree very much; 5 = agree very much). The measure yields an overall Machiavellianism score that is the average of the scores for the 20 items.

The translation of the original scale was completed by two researchers separately. Consensus about the proper translation of each item was reached between them. A third translator then checked if the translations agreed with the original items. The test–retest reliability of this new translation was tested on an independent sample of 44 7th graders from one of the schools involved in this project but who were not part of the actual study. They completed the measure twice across 9 weeks in the fall of the school year. The test–retest correlation of the overall Machiavellianism score in this sample was .62. Cronbach’s α of the scale was .76 at Time 1 and .70 at Time 2. Thus, the psychometric properties of the scale were acceptable for inclusion in the actual study. The internal consistency reliability of the 20-item scale in the total sample of 806 participants in this study was .61 (Cronbach’s α).

Social Intelligence

Social intelligence was measured with a translation of the Tromsø Social Intelligence Scale (Silvera et al. 2001). The Dutch translation was used successfully in a recent study by Meijs et al. (in press). The scale consists of 21 items (e.g., “I can predict other people’s behavior”), that are scored on a 7-point scale (1 = does not describe me at all; 7 = describes me very well). The internal reliability of the scale was found to be satisfactory (Cronbach’s α = .78).

Identification of Bullies

Participants who scored in the top 15% of the distribution of the standardized bullying score (standardized within classrooms and gender) were identified as bullies. This is a relatively strict criterion but has been used more often to identify bullies in previous studies (see e.g., Camodeca et al. 2003), and is also otherwise a common criterion to identify subgroups in peer relations research using sociometric measures (see Cillessen 2009). As a result, 120 of the 806 participants (14.9%) were identified as bullies. This method yielded a bully sample of 52 boys with an average standardized bully score of 2.04 (SD = .84), and a bully sample of 68 girls with an average standardized bully score of 1.79 (SD = 1.49). For comparison purposes, 25 boys and 25 girls were randomly selected from the remainder of the sample to form the non-bully comparison group.

Results

Analysis Strategy

The analyses of this paper proceeded in two steps. First, cluster analysis was conducted on the subsamples of male and female bullies to identify subtypes. Second, the resulting clusters were examined on a number of new variables, to validate their nature.

Table 1 shows the means and standard deviations for bullying, the clustering variables, and the profiling variables by gender in the analysis sample (N = 170). Gender comparisons indicated that boys scored significantly higher than girls on peer- and self-reported bullying, physical and verbal aggression, rejection, leadership, and Machiavellianism. Girls scored higher than boys on social network centrality.
Table 1

Descriptive statistics of study variables in the analysis sample (N = 170)

 

Boys (N = 77)

Girls (N = 93)

Gender difference

M

SD

M

SD

t

p

Social intelligence

4.80

.60

4.90

.53

−1.14

.261

Machiavellianism

2.58

.44

2.43

.33

2.47

.015

Relational aggression

−.75

1.32

.91

1.27

−.80

.424

Perceived popularity

.71

1.05

.46

.84

1.69

.094

Bullying

1.84

1.48

.71

1.09

5.57

.001

Physical aggression

1.41

1.44

−.13

.84

8.29

.001

Verbal aggression

1.60

1.45

.39

1.21

5.83

.001

Social rejection

.73

1.22

.12

.95

3.58

.001

Leadership

1.03

1.47

.46

1.25

2.69

.008

Network centrality

3.63

2.31

4.78

2.80

−2.93

.004

Self-reported bullying

9.41

3.59

8.13

2.39

2.68

.008

Means that are italicized were significantly different by gender

Table 2 shows the intercorrelations between the same variables by gender in the analysis sample. The correlations in general followed a meaningful pattern that is common for these constructs in other research. Notice in particular that, for both genders, all measures of aggression were positively correlated with each other, as well as with bullying. All correlations were compared by gender using Fisher’s r-to-z transformations. As can be seen in Table 2, a number of correlations were significantly different by gender. In particular, the association between bullying and physical aggression was stronger for boys than for girls.
Table 2

Correlations between study variables in the analysis sample (N = 170)

 

1

2

3

4

5

6

7

8

9

10

11

1. Social intelligence

 

.05

.35*

.30*

.24*

.07

.28*

.12

.38*

.11

.17

2. Machiavellianism

−.56*

 

.24*

.19

.19

.12

.19

.21*

.22*

.06

.15

3. Relational aggression

.23*

−.20

 

.55*

.73*

.28*

.66*

.39*

.69*

.02

.34*

4. Perceived popularity

.39*

−.19

.39*

 

.50*

.02

.31*

.01

.64*

.33*

.33*

5. Bullying

.09

.02

.66*

.45*

 

.28*

.74*

.49*

.65*

−.07

.13

6. Physical aggression

.01

−.13

.44*

.30*

.55*

 

.54*

.39*

.25*

−.28*

−.01

7. Verbal aggression

.15

−.05

.71*

.47*

.85*

.67*

 

.49*

.56*

−.13

.11

8. Social rejection

−.14

.09

.54*

.03

.42*

.41*

.51*

 

.23*

−.19

.02

9. Leadership

.26*

−.16

.66*

.65*

.59*

.54*

.62*

.32*

 

.03

.24*

10. Network centrality

.26*

−.02

.16

.46*

.14

.07

.16

−.02

.23*

 

.27*

11. Self-report bullying

−.10

.25*

.10

.10

.32*

.37*

.35*

.32*

.17

.09

 

Correlations italicized were significantly different by gender. Girls above the diagonal, boys below the diagonal

p < .05

Cluster Analysis

Based on the considerations given above, the clustering variables were perceived popularity and relational aggression (from peer nominations), and social intelligence and Machiavellianism (from self-reports). A hierarchical cluster procedure was used, starting with one cluster containing all of the observations. Similar to previous studies in this field (e.g., Cillessen et al. 1992; Estell et al. 2002), Ward’s clustering method based on squared Euclidian distances between cases on each clustering variable was used. Ward’s method minimizes the variance within clusters and maximizes between-group variance (Estell et al. 2002; Hair and Black 2000).

For boys, a three-cluster solution was identified as the most optimal solution to differentiate the boys on the four clustering variables. The first cluster, including 14 boys, was set apart from the remaining bullies based on social intelligence and popularity. This cluster was significantly more socially intelligent than the rest, M’s = 5.42 versus 4.68, F(1, 50) = 15.37, p < .001, and significantly more popular, M’s = 1.94 versus .72, F(1, 50) = 39.03, p < .001. The second cluster, including 28 boys, was differentiated from the remaining boys based on relational aggression. This group was significantly less relationally aggressive then the rest, M’s = .17 versus 2.15, F(1, 50) = 118.77, p < .001. The third cluster, including 10 boys, differed significantly from the first two (see below), but not on a single variable.

The clustering process for girls also generated a three-cluster solution. The first cluster, including 31 girls, was set apart from the remaining girls based on relational aggression. The girls in this cluster were significantly more relationally aggressive than the remaining girls, M’s = 2.21 versus .42, F(1, 66) = 180.14, p < .001. The second cluster, including 20 girls, significantly differed from the other clusters (see below), but not on a single variable. The remaining 17 girls had significantly lower scores on social intelligence than the other girls, M’s = 4.58 versus 5.07, F(1, 66) = 14.90, p < .001.

The means and standard deviations of the clusters on the four clustering variables are shown in Table 3. To further identify each cluster, two analyses were conducted. First, the three clusters were compared to each other on each clustering variable in a three (Cluster) ANOVA. To examine if the clusters differed in the extremity of their bullying, this ANOVA was also run on the original bullying score. Significant cluster effects were followed up with a Tukey post-hoc comparison. Second, each cluster was compared with the non-bully comparison group in a two (Group: Cluster vs. Comparison) ANOVA. The results of these analyses are also shown in Table 3.
Table 3

Comparisons of bullying subtypes on clustering variables

 

Cluster 1

Cluster 2

Cluster 3

Non-bullies

 

Boys

n = 14

n = 28

n = 10

n = 25

F(3, 73)

 Social intelligence

5.42a (.52)

4.74b (.66)

4.51b (.52)

4.66 (.36)

8.17*

 Machiavellianism

2.30a (.36)

2.73b (.47)

2.56ab (.45)

2.58 (.39)

4.39*

 Relational aggression

2.03a (.62)

.17b (.70)

2.31a (.54)

−.16 (.81)

60.00*

 Perceived popularity

1.94a (.24)

.85b (.78)

.36b (.30)

.01 (1.14)

23.38*

Girls

n = 31

n = 20

n = 17

n = 25

F(3, 89)

 Social intelligence

5.13a (.35)

4.99a (.55)

4.58b (.49)

4.78 (.59)

7.99*

 Machiavellianism

2.56a (.35)

2.46ab (.28)

2.27b (.34)

2.36 (.31)

4.17*

 Relational aggression

2.21a (.50)

.77b (.33)

.01c (.56)

−.15 .79)

132.66*

 Perceived popularity

1.07a (.69)

.77a (.52)

−.22b (.62)

−.07 (.70)

23.89*

Means within rows that do not share a subscript were significantly different between clusters. Cluster means that are italicized were significantly different from the non-bully comparison group. Standard deviations are between parentheses

p < .05

The results for boys showed that Cluster 1 bullies were socially intelligent and relationally aggressive. They were perceived as popular by their classmates and scored low on Machiavellianism. The differences with other bullies and non-bullies were most evident for social intelligence and popularity. Therefore, this group was labeled popular-socially intelligent. Cluster 2 bullies were relatively popular and had average scores for social intelligence, relational aggression, and Machiavellianism. They only differed from non-bullies in terms of popularity. Therefore, they were labeled popular moderate. Cluster 3 bullies were considerably relationally aggressive and had average Machiavellian scores. Compared to other bullies and non-bullies they were less popular and socially intelligent. The differences with non-bullies and other bullies were most obvious for popularity and social intelligence. Therefore, this group was labeled low popular-low socially intelligent.

For girls, Cluster 1 bullies were socially intelligent and above average Machiavellian beliefs. They were popular and relationally aggressive. Differences with other bullies and non-bullies were evident for popularity, social intelligence, and relational aggression. Given the correspondence of this cluster to the first cluster for boys, it was also called popular-socially intelligent. Cluster 2 bullies were moderately popular and above average in social intelligence. They scored average on Machiavellianism and high on relational aggression. Because of their generally moderate scores on the clustering variables, they were called popular moderate. Cluster 3 bullies were unpopular and below average in Machiavellianism. They scored significantly below other bullies on relational aggression and social intelligence. Therefore, they were called low popular-low socially intelligent.

Cluster Profiling

To further validate the clusters, they were compared on other variables that had not been used to identify them. Therefore, a three (Cluster) ANOVA was conducted for each gender on the scores for physical aggression, verbal aggression, social rejection, leadership, network centrality, and self-reported bullying. Significant cluster effects were followed by Tukey post-hoc comparisons. Each cluster was also compared to the non-bully comparison group in a two (Group: Cluster vs. Comparison) ANOVA. Table 4 presents the results of these analyses.
Table 4

Comparisons of bullying subtypes on validation variables

 

Cluster 1

Cluster 2

Cluster 3

Non-bullies

 

Boys

n = 14

n = 28

n = 10

n = 25

F(3, 73)

 Verbal aggression

2.85a (1.20)

1.66b (.92)

2.96a (.62)

.28 (1.17)

10.69*

 Physical aggression

2.20 (1.30)

1.54 (1.21)

1.95 (1.81)

.56 (1.30)

1.18

 Social rejection

.96ab (1.35)

.43a (.98)

2.06b (.99)

.37 (1.17)

8.29*

 Leadership

2.78a (.93)

.72b (1.21)

1.58b (1.38)

.13 (1.07)

14.37*

 Network centrality

5.32a (1.96)

3.72ab (2.14)

3.17b (2.02)

2.93 (2.34)

3.89*

 Self-reported bullying

9.57 (3.32)

10.26 (3.95)

10.50 (2.72)

7.92 (3.32)

.24

Girls

n = 31

n = 20

n = 17

n = 25

F(3, 89)

 Verbal aggression

1.22a (1.42)

.01b (.65)

.17b (1.16)

−.19 (.69)

7.90*

 Physical aggression

.01 (1.12)

−.18 (.69)

−.02 (.97)

−.35 (.27)

.26

 Social rejection

.49a (.94)

−.20b (.50)

.19ab (.94)

−.15 (1.11)

4.190*

 Leadership

1.61a (1.31)

.36b (.76)

−.32b (.54)

−.36 (.58)

21.79*

 Network centrality

4.78 (2.94)

5.92 (2.44)

4.08 (2.76)

4.37 (2.78)

2.14

 Self-reported bullying

8.87a (2.28)

9.25a (2.73)

7.18b (1.85)

6.69 (1.84)

4.16*

Means within rows that do not share a subscript were significantly different between clusters. Cluster means that are italicized were significantly different from the non-bully comparison group. Standard deviations are between parentheses

p < .05

For boys, popular-socially intelligent bullies scored significantly higher than other bullies on leadership. They also scored significantly higher than low popular-low socially intelligent bullies on network centrality. They scored significantly higher than non-bullies on physical aggression, verbal aggression, leadership, social rejection, and network centrality.

Popular-moderate bullies scored significantly lower on the aggression measures than other bullies, however, compared to non-bullies they were more aggressive. They scored significantly lower than popular-socially intelligent bullies on network centrality, social rejection, and leadership, but their centrality and rejection scores did not differ from non-bullies. Low popular-low socially intelligent bullies looked like popular-socially intelligent bullies, with the exception of network centrality, social rejection, and leadership. For network centrality, they did not differ from non-bullies, but for leadership and social rejection, they scored significantly higher than non-bullies. They scored significantly lower than popular-socially intelligent bullies on leadership and network centrality.

For girls, popular-socially intelligent bullies were more verbally aggressive than other bullies. They also scored significantly higher on leadership and social rejection. They were more physically aggressive than non-bullies. Popular-moderate bullies were more central members of the peer group than non-bullies. They differed significantly from non-bullies on all measures except verbal aggression and rejection. Nevertheless, their mean scores on the five profiling variables did not distinguish them from the other bullies. Low popular-low socially intelligent bullies were physically and verbally more aggressive than non-bullies. They did not differ from other bullies or non-bullies in social rejection. They scored lower on self-reported bullying than the other two groups. One would expect them to score higher on self-reported bullying than non-bullies, but this difference was not significant.

Discussion

This study confirmed the need to differentiate between bullies instead of assuming that bullies are a homogeneous group. The results supported Sutton et al.’s (1999a) hypothesis that at least some bullies are socially intelligent and have the ability to manipulate the peer group. At the same time, the results also indicated that not all bullies can be typified as socially intelligent and well-accepted peer group members. This is in agreement with Crick and Dodge’s (1994) assumption that aversive behavior such as bullying is the result of incompetencies or deficiencies, such as a malfunctioning social information processing system. The current study demonstrated significant diversity among bullies in the degree of social intelligence and status. A similar bullying profile was found for both genders that included in both cases a socially intelligent and powerful group and a less socially intelligent and unpopular group.

It was hypothesized that the social function of bullying differs between bullies and is associated with status, skills, and position in the peer network. Different functions of bullying will demand different skills and levels of social influence. Indeed, the results suggest that the identified profiles of bullies reflect variation in the social function of bullying. Our analyses showed that high social status is accompanied by social intelligence and the use of relational aggression. The bullies with this set of characteristics, the popular-socially intelligent bullies, may use their skills to gain dominance. Their centrality in the group would enable them to persuade others to ignore the victim and believe their backbiting. In this case, bullying is used to acquire power and influence.

Our findings are consistent with previous findings of strong associations between relational aggression, perceived popularity, and social intelligence. This suggests that these three concepts are unified, and that popularity and social intelligence may be preconditions for the use of relational aggression (Björkqvist et al. 1992; Salmivalli et al. 2000). Putallaz et al. (2007) suggested, that the indirect character of relational aggression demands social intelligence to cover the aggressive acts. Surprisingly, we found that male bullies who were unpopular and lacking social intelligence were also highly relationally aggressive. Thus, at least for boys, relational aggression may not always require social intelligence.

Findings by Xie et al. (2002) may explain this exceptional finding. They distinguished social aggression from direct relational aggression. Social aggression referred to covert aggressive attacks, such as gossiping, while direct relational aggression referred to more visible attacks, such as ignoring or excluding others. The respondents in their study saw social aggression as provoking behavior causing the most damage, while direct relational aggression was perceived as reactive. It is perhaps the concealed nature of social aggression that demands social intelligence, which may explain why unpopular-low socially intelligent bullies were seen as relationally aggressive but lacking social intelligence.

The results reported by Xie et al. (2002) may also contribute to typifying this last group as socially deficient, in agreement with Crick and Dodge (1999). They hypothesized that bullies attribute hostile intent to ambiguous cues and therefore respond with excessive aggression. Unpopular-less socially intelligent bullies may use more direct relational aggression than popular-socially intelligent bullies, who may use more social aggression. The reactive nature of direct relational aggression as proposed by Xie et al. (2002) is consistent with the perspective of the bully proposed by Crick and Dodge (1999) and the unpopular-socially less intelligent group of bullies in this study.

The social functions of bullying by unpopular-low socially intelligent bullies may be more intuitive or automatic. Perceiving ambiguous cues as hostile may reflectively trigger defending behavior. These bullies may respond with aggression because they feel threatened. Their aggression may be more reactive, and thus have a different function than the bullying of popular-socially intelligent bullies that may be more instrumental to gain dominance.

While for boys the less popular and less socially intelligent bullies were the most rejected ones, for girls social rejection was associated with high social intelligence and status. Contrary to boys, the popular-socially skilled female bullies were the most disliked. They scored significantly higher on rejection than moderate-popular bullies and non-bullies.

Before explaining the gender differences, we emphasize that the received dislike nominations are a measure for social rejection (Coie et al. 1982). Because the cluster analyses require continuous scores, the standardized dislike nominations were used as a measure of social rejection instead of the categorical status types. However, the sociometric status type “rejected” is a two-dimensional construct based on both liked and disliked nominations. It is therefore different from the continuous measure used here. It is possible that the bullies in our sample received not only disliked nominations but also liked nominations and, as a result, would be classified as sociometrically “controversial.”

The gender differences in the social rejection of bully types may be explained by gender differences in relationships with peers more generally. They may explain why, contrary to their male counterparts, popular-socially skilled female bullies scored significantly higher on social rejection than moderate-popular bullies and non-bullies. Eder (1985) introduced the “cycle of popularity” to describe how disliking and popularity in girls go hand in hand. Indeed, recent findings support the assumption that high status does not exclude the possibility of social rejection (Caravita et al. 2008; Estell et al. 2007). The cycle of popularity (Eder 1985) addressed the role of friendships with popular girls in adolescence. Being friends with popular girls is important for one’s own status, however, popular girls tend to avoid affiliations with lower-status girls. Because they reject the offers of friendship from lower-status peers, popular girls are also the most disliked girls. Consistent with Eder’s cycle of popularity (1985), the most popular girls in our study were also the most disliked ones.

Cillessen and Borch (2006) examined the development trajectories of sociometric and perceived popularity and the associations with aggression. The effect of relational aggression on perceived popularity changed over time. Cillessen and Mayeux (2004) also discussed the changing association between aggression and perceived popularity across development. These studies indicate that these group constructs are sensitive to developmental changes. It is possible that the different social functions of bullying are a reflection of developmental changes, and that the different subtypes of bullies are in fact adolescents at different stages of social cognitive and behavioral development.

Furthermore, the different types of bullies may also represent different responses to normal social contextual changes taking place in adolescence. Pellegrini and Long (2002) indicated that the transition from primary to secondary school, and the disruptions of peer relationships that come with it, may increase bullying behavior depending on adolescents’ needs to establish centrality in a peer group. If being central in a peer network is important, bullying may be an efficient (but undesirable) way to accomplish it. Individual differences between adolescents in this need may also influence their bully subtype classification.

In summary, our results highlight the significance of differentiating between bullies, and identifying unique profiles of bully types. Gender appears to play a role in determining these profiles. Most importantly, bully types can be described by the different social functions or provisions that bullying may have for each of them. Thus, our classification of bullies primarily reflects a functional approach to bullying. Underlying functions, combined with available social-cognitive and behavioral skills, and afforded by a certain position in the peer group, may interact to determine an adolescent male of female’s bullying classification. The functional approach also has to be a developmental approach, as functions, skills, and peer group structures change across developmental time. Changes and transitions during adolescence are associated with bullying behavior, and these developmental changes may also influence the social functions and provisions of bullying behavior.

Limitations and Directions for Future Research

Although we found significant diversity in behavior, status, and skills among bullies, differences with non-bullies were not always obvious. The clustering variables in our study clearly showed their usefulness in differentiating among bullies; however, some of them did not distinguish all types of bullies from non-bullies. Therefore, it may be useful to include existing scales, such as the Participants Roles measure (Salmivalli et al. 1996), in a future study. Separating non-bullies who help the victim from those who assist or reinforce the bully may lead to a clearer distinction between non-bullies who have a pro-bullying attitude and those who have an anti-bullying attitude. This may change the differentiation of bullies from non-bullies. Currently, the non-bullies were a random sample of all non-bullies existing in the larger normative sample, ignoring any differentiation that may exist among them. Nevertheless, our main goal was to identify bully types and the clustering variables used in our study showed their relevance for this goal.

Our current study also did not examine the possible influences of classroom or school norms. Multilevel analysis would be required to examine the influence of classroom or school variables on the associations among the clustering and profiling variables. However, the effects of classrooms or schools may have been small as the two schools in this study were quite similar and the overall sample rather homogenous in many ways. Nevertheless, the issue of school or classroom contextual effects is an important one that has been generally undervalued in peer relations research and deserves more attention in future research.

Another limitation is the non-inferential and descriptive nature of cluster analysis (Hair and Black 2000). Cluster analysis requires few statistical assumptions and will always create a cluster solution. Some groups will always be found. Nevertheless, valid groups are not always found and the groups identified in our study clearly had validity. Thus, despite the inherent exploratory nature of cluster analysis, we believe in its value in the current study, as also demonstrated in previous peer relations projects (e.g., Cillessen et al. 1992; Estell et al. 2002; French 1990; Salmivalli and Kaukiainen 2004).

A final limitation is that we only measured eighth graders at one point in time. Inclusion of other age groups and longitudinal designs could identify developmental changes in bullying classifications and perhaps reveal variation in bully subtypes across development. This would be consistent with the developmental functional approach advocated above.

Still to be explored are the motives for bullying. The current study identified types of bullies and their behavioral, social-cognitive, and status profiles. Motivation was not considered. Yet, there may be important differences between bully subtypes in underlying motives. In a recent study, LaFontana and Cillessen (in press) documented developmental changes in children and adolescents’ needs to high status and popularity. Coincidentally, the peak of this need appears to be somewhere around the age of the participants of the current study. Yet, there also appear to be large individual differences in this need. Thus, examining the degree to which adolescents prioritize peer status and how this is related to bullying aggression seems an important avenue for follow-up research.

The findings of this study have implications for the design of anti-bullying programs. A single “one size fits all” intervention strategy for reducing bullying behavior may not be sufficient. It may be of crucial importance to adjust bullying intervention programs to the bullying subtypes one is working with. For some bullying situations it may be necessary to target the whole peer network, for instance by means of changing the attitudes of bystanders (Salmivalli et al. 2005) or restructuring classroom networks as suggested by Salmivalli (1999). Because socially intelligent bullies have the skills to interact and use the social structure to harass victims, group interventions may be the best choice to reduce their bullying. Individual interventions, such as self-reflection or feedback about social behavior (Salmivalli 1999), may be more appropriate for bullies who are less well-connected and whose bullying is more driven by individual forces rather than carried by the structure of the peer network.

This study uniquely contributes to our understanding of bullying behavior. The bully profiles that were identified and the differences among them suggest that bullying in adolescence may serve a variety of social functions that are not a homogenous set. A framework was also provided to relate the various social functions of bullying to developmental changes in adolescence. Recognizing the cross-sectional functional heterogeneity of bullying, in conjunction with its longitudinal developmental variability may be the next big challenge in bullying research.

Acknowledgments

This research was supported by a Master’s Research Grant from the Behavioural Science Institute to the first author. The authors are grateful to the students who participated in this study. Special thanks are also due to the teachers and administrators of the Valuas College, Venlo and the BBC College, Panningen, The Netherlands who made this research possible.

Copyright information

© Springer Science+Business Media, LLC 2009