Bully-Victim Dyads - 1 -
Running head: BULLY-VICTIM DYADS
The dyadic matches of bullying and victimization: Testing a dual perspective theory
University of Groningen, the Netherlands
University of Groningen, the Netherlands
Bonne J.H. Zijlstra
University of Amsterdam, the Netherlands
Andrea F. De Winter
University Medical Center Groningen, University of Groningen, the Netherlands
Frank C. Verhulst
Erasmus Medical Center, Rotterdam, the Netherlands
University Medical Center Groningen, University of Groningen, the Netherlands
René Veenstra and Siegwart Lindenberg, Department of Sociology and Interuniversity Center
for Social Science Theory and Methodology (ICS); Bonne J.H. Zijlstra, Department of Educational
Sciences; Andrea F. De Winter and Johan Ormel, Department of Psychiatry, Graduate School of
Behavioural and Cognitive Neurosciences, and Graduate School for Experimental Psychopathology;
Frank C. Verhulst, Department of Child and Adolescent Psychiatry.
This research is part of the TRacking Adolescents’ Individual Lives Survey (TRAILS).
Participating centers of TRAILS include various Departments of the University of Groningen, the
Erasmus Medical Center of Rotterdam, the Radboud University of Nijmegen, University of Utrecht,
and the Trimbos Institute the Netherlands. TRAILS is financially supported by grants from the
Netherlands Organization for Scientific Research (GB-MW 940-38-011, GB-MAGW 480-01-006,
Bully-Victim Dyads - 2 -
GB-MAGW 457-03-018, NWO 175.010.2003.005, ZonMw 100-001-001, ZonMw 60-60600-98-
018), the Ministry of Justice (WODC), and by the participating centers. We are grateful to Menno
Reijneveld for useful comments on an earlier version of this article. Earlier versions of this paper
were presented at the biennial meeting of the Society for Research on Adolescence, San Francisco,
CA, United States of America, March, 2006, at a NSF-Sponsored Conference on Modeling
Interdependent Data in Developmental Psychology, Lawrence, KA, June, 2006, and at the biennial
meeting of the International Society for the Study of Behavioral Development, Melbourne, Australia,
Correspondence should be addressed to the first author at the ICS, University of Groningen,
Grote Rozenstraat 31, 9712 TG Groningen, the Netherlands. Phone: +31-50-3636240; FAX: +31-50-
3636226. E-mail: firstname.lastname@example.org; http:/www/gmw.rug.nl/~veenstra
Bully-Victim Dyads - 3 -
Many researchers have argued for a conceptualization of the dyad as the unit of analysis to
understand relationships among classmates, but there has been little response so far. We identified
the characteristics of bully-victim matches using nominations of who bullies who and who is
victimized by whom and analyzing binary network data with nominator, target, and dyadic covariates,
and random effects (p2 model). We viewed bullying and victimization separately from the point of
view of the bully and the victim using a goal-framing approach to predict the characteristics of the
match. The two perspectives were highly complementary. The probability of a bully-victim
relationship is higher if the bully is more dominant than the victim, and if the victim is more
vulnerable than the bully and more rejected by the class. In a bully-victim dyad, boys were more
often the bullies. We did not find a sex effect for victimization. Liking reduced and disliking
increased the probability of a bully-victim relationship.
KEY WORDS: bullying, peer relations, social networks, victimization, elementary school students
Bully-Victim Dyads - 4 -
The dyadic matches of bullying and victimization: Testing a dual perspective theory
Word count: Abstract 160, Text 6794, References 1737, Tables 437
Bullying and victimization have been important subjects in developmental studies. However,
there is still a lack of dyadic studies of bullying and victimization. Despite recognition that bullying
occurs disproportionately in specific dyads (Coie et al., 1999; Olweus, 1978), not much is known
about the specific matches. Children are routinely classified as bullies and victims, but rarely is it
known which bullies harass which victims (Rodkin & Berger, 2005). Some children have a tendency
to be involved in bullying, as a bully, a victim, or a bully-victim (Schwartz, 2000), but the question
is, with whom? Matches come from two sides. Do self-proclaimed bullies have typical kinds of
victims? Do self-proclaimed victims have typical kinds of persons who they identify as bullies?
To answer the above questions, it is necessary to turn to network analysis, and so far use of
this approach has been scarce in the study of bullying and victimization. Once bullying and
victimization are considered from a dyadic match point of view, another possibility arises. It is
possible that the process that leads adolescents to bully certain others differs from the process that
leads victims to feel bullied. The subjective experience of bullying and of being bullied may not
match. This possibility suggests use of a dual perspective approach that looks at bullying from the
points of view of both the self-proclaimed bully and the self-proclaimed victim. This may have an
added advantage of contributing to a better understanding of the associations of bullying with other
peer adversities, such as rejection and isolation (see also: Salmivalli & Isaacs, 2005). The main goal
of the current study was to formulate and test such a dual perspective theory of bullying and
victimization and test it using network analyses that identify dyadic matches, i.e., the covariates of
the nominator (i.e., the informer) and the target (i.e., the person nominated as bully or victim).
With our dyadic match approach, it was not our aim to deny the importance of the group as a
Bully-Victim Dyads - 5 -
context of bullying (Whitney & Smith, 1993; Salmivalli, Lagerspetz, Björkqvist, Osterman, &
Kaukiainen, 1996; Atlas & Pepler, 1998), but we believe that, for an understanding of bully-victim
relations in a peer context, much can be learned from dyadic matches.
A Dual Perspective Theory of Bullying
In order to get a better grip on the dyadic nature of the bully-victim relationship, it seems
plausible to consider the possibility that the subjective experiences of bullying and being bullied do
not need to match. Teasers and their targets often have different intentions and perceptions of an act
(Shapiro, Baumeister, & Kessler, 1991; Kowalski, 2000). When bullying is viewed in terms of
dyadic matches, two kinds of dyadic information are obtained: from the self-proclaimed bully (‘Who
do you bully?’) and from the self-proclaimed victim (‘By whom are you bullied?’). In both cases, the
goals pursued should be important for explaining the specific characteristics of a ‘match’.
To predict who bullies who and why, we used a goal-framing approach (Lindenberg, 2006), in
which it is not only the substantive content of a goal that is crucial for action, but also that an
activated goal will make the person sensitive to opportunities for its realization (Klinger, 1975).
Conversely, the goal may become activated merely by exposure to an opportunity to realize it (Shah
& Kruglanski, 2003). In the literature on victimization, this insight has long been used (see, for
example, Miethe & McDowall, 1993). The stronger the goal, the more likely that it will create this
double sensitivity: a keen awareness of opportunities to realize the activated goal, and a readiness to
have the goal activated by detecting opportunities to realize it. What kinds of characteristics do
potential (self-proclaimed) bullies have and what characteristics are they sensitive to? What
characteristics do (self-proclaimed) victims have and what do they particularly focus on? Answers to
these questions yielded testable hypotheses about bully-victim matches from the points of view of
both the self-proclaimed bully and the self-proclaimed victim. The hypotheses were derived from
assumptions about goals of potential bullies and victims that relate to bullying and being bullied,
even if these goals and their consequences for sensitivities were not directly assessed.
The bullying relationship has been called asymmetric by definition (Olweus, 1993). Olweus
stated that the term bullying should not be used when two children of the same strength are fighting
Bully-Victim Dyads - 6 -
or quarrelling. It is likely that for youth, the term 'bullying' also implies a power imbalance, referring
to status goals for potential bullies and protective goals for potential victims (Salmivalli, 2001). We
focused on the points of view of the self-proclaimed bully and the self-proclaimed victim. We
purposely did not consider the category of children nominated as both bully and victim.
The point of view of the bully Hawley (1999) posits that children bully weaker children to gain,
among other things, higher status among peers. In addition, status striving has been identified as one
of the ubiquitous human goals (Huberman, Loch, & Önçüler, 2004; Lindenberg, 2001). It seems
warranted, therefore, to take striving to improve one's status as the major substantive goal for the
self-proclaimed bully. There are two aspects to the status goal (see also: Whiting & Edwards, 1973).
On the one hand, there is a need for domination, and on the other hand, there is a need to get social
approval for being special in comparison to others. Research by Vailliancourt, Hymel, and
McDougall (2003) has shown that both power differences (referring to children who have power over
others, who can pressure others into doing things) and status differences (referring to children who
are most popular, most liked, and least disliked in a school grade) are key aspects of bullying. If the
domination component is relatively strong (in comparison to the social approval component), then it
is likely that children bully to experience and show their domination over other children.
A number of points follow from the goal-framing approach for the characteristics of the self-
proclaimed bully and the characteristics of the victim they are typically matched with. First, children
with a high score on dominant aggressiveness are more likely to have the goal to be dominant as a
focal goal than children with a low score (compare Hawley, 2003; Rodkin, Farmer, Pearl, & Van
Acker, 2000; LaFontana & Cillessen, 2002). Because boys are generally more aggressive than girls,
self-proclaimed bullies are more likely to be boys than girls, and they are likely to be high on
dominant aggressiveness. This is also supported by the finding that bully-victim relationships are
dyads in which boys are usually the aggressors and boys or girls the victims (Espelage, Mebane, &
Adams, 2004; Hanish & Guerra, 2004; Klicpera & Gasteiger Klicpera, 1996; Pellegrini, Bartini, &
Brooks, 1999; Schwartz, 2000; Vermande, Van den Oord, Goudena, & Rispens, 2000; Salmivalli,
2001). Second, potential bullies do not belong to the rejected group and are even accepted to some
Bully-Victim Dyads - 7 -
degree because social approval is part of their status striving. Third, potential bullies have a keen eye
for victims that help them realize their dual status goals. This means that potential bullies are likely to
spot children who are vulnerable, that is, who offer a high probability of letting themselves be
dominated. Because bullies are also not likely to trade social approval for domination if both can be
had simultaneously (at times, they may even get social approval from bystanders (O'Connell, Pepler,
& Craig, 1999), we expected them to be keenly aware of other children's lack of social support (i.e.,
their being rejected by other children). In short, having the goals to be dominant and to get social
approval, the potential bully spots who is vulnerable and at the same time is not supported by others.
Thus, from the point of view of the bully, our bully-victim profile hypotheses were the following:
self-proclaimed bullies are likely to be (a) boys, (b) dominantly aggressive, and (c) accepted; and
they are matched with victims that are likely to be (d) vulnerable, (e) rejected, and (f) not aggressive.
Notice that the power imbalance does not apply to ‘being accepted’ because that has to do with the
bully's goal to get social approval.
The point of view of the victim The goals of the victims are likely to be different. Children who
feel vulnerable (i.e., feel easily hurt by others, can't make others listen to them, feel isolated) are
more likely than children who feel less vulnerable to have the goal to avoid being hurt as a focal goal
(compare Juvonen & Graham, 2001; Perry, Williard, & Perry, 1990; Olweus, 1978). Such children
are likely to be aware of other children who are more aggressive and less vulnerable than they are
and who are, therefore, a potential threat. They may view interpersonal situations as stressful and
anxiety producing. Avoidance evoked by fear or wariness may be the result (Burgess, Wojslawowicz,
Rubin, Rose-Krasnor, & Booth-Laforce, 2006). This may even signal their vulnerability to those who
are at least moderately interested in domination (Boldizar, Perry, & Perry, 1989; Salmivalli & Isaacs,
2005), thereby triggering dominant behavior also in classmates who would otherwise not show it.
Because of the ‘low power’ situation of self-proclaimed victims, it is likely that status goals are more
or less inhibited, which makes the goal not to get hurt even more prominent (Keltner, Gruenfeld, &
Anderson, 2003). Finally, because boys are generally more dominantly aggressive than girls, self-
proclaimed victims can be expected mostly to be matched to male bullies. In sum, from the point of
Bully-Victim Dyads - 8 -
view of the victim, our victim-bully profile hypotheses were the following: self-proclaimed victims
are likely to be (a) vulnerable, (b) rejected, and (c) not aggressive; and they are matched with bullies
who are likely to be (d) boys, (e) dominantly aggressive, and (f) not vulnerable.
As can be seen, the hypotheses about the profiles of bully-victim matches are almost identical
from both points of view. Only ‘being accepted’ is not part of the match but part of the predicted
characteristics of the self-proclaimed bully. If ‘being accepted’ also showed up in the profile of the
bully as identified by the self-proclaimed victim, this would provide extra evidence that the
perspective of the self-proclaimed bully and that of the self-proclaimed victim match.
Relational hypotheses (from both points of view)
It has been found that bullying is more often directed toward children of the same sex than
toward children of the opposite sex (Pellegrini & Long, 2002). Rodkin and Berger (2005) made clear
that boys who harass boys receive even more approval or prestige than boys that harass girls.
Together with the assumption that boys are more often self-proclaimed bullies than girls (see above),
this led us to expect that boy-boy bully-victim relationships would be more likely than girl-girl or
mixed-sex pairs of actors.
With regard to liking and disliking, we expected that liking a person would reduce both an
existing desire to dominate this person and the likelihood of reporting that one is being bullied by the
liked person (Keltner, Young, Heerey, Oemig, & Monarch, 1998). Conversely, we expected that
disliking a person would increase both the existing desire to dominate this person and the likelihood
of reporting that one is being bullied by the disliked person (see also Peets, Hodges, & Salmivalli,
The Present Study
Survey instruments that are responsive to relational contexts have been considered important
means of understanding bully-victim processes (Coie et al., 1999; Rodkin & Berger, 2005;
Salmivalli, 2001). We adopted a new technique for examining the questions of who bullies who and
who is victimized by whom. Bullying and victimization were assessed using peer nominations, in
Bully-Victim Dyads - 9 -
which children nominated who they bullied and by whom they were bullied. We also adopted a new
statistical model: the p2 model (Van Duijn, Snijders, & Zijlstra, 2005; Zijlstra, Van Duijn, &
Snijders, 2006), a model for the analysis of binary network data with nominator, target, and dyadic
covariates, and random effects.
The participants were 918 Dutch preadolescents who took part in a sociometric data
collection, a part of the TRacking Adolescents’ Individual Lives Survey (TRAILS). TRAILS is a
prospective cohort study of Dutch preadolescents who will be measured biennially until they are at
least 25 years old.
The present study was part of the first assessment wave of TRAILS, which ran from March
2001 to July 2002. TRAILS is designed to chart and explain the development of mental health and
social development from preadolescence into adulthood. The TRAILS target sample consisted of pre-
adolescents living in five municipalities in the north of the Netherlands, including both urban and
rural areas (De Winter et al., 2005; Oldehinkel, Hartman, De Winter, Veenstra, & Ormel, 2004).
Of the children approached for enrollment in the study (selected by the municipalities and attending a
school that was willing to participate; N = 3145 children from 122 schools, response of schools 90.4
percent), 6.7 percent were excluded because of incapability or language problems. Of the remaining
2935 children, 76.0 percent were enrolled in the study, yielding N = 2230 (consent to participate:
both child and parent agreed). No non-response bias was found in our study for the estimation of the
prevalence rates of psychopathology, including antisocial behavior. Boys, children from lower social
strata, and children with worse school performance were somewhat more likely to belong to the non-
response group (De Winter et al., 2005).
Well-trained interviewers visited one of the parents (preferably the mother, 95.6 percent) at
home to administer an interview covering a wide range of topics, including the child’s developmental
history and somatic health, parental psychopathology, and care utilization. The parent was also asked
Bully-Victim Dyads - 10 -
to fill out a questionnaire. Children filled out questionnaires at school, in class, under the supervision
of one or more TRAILS assistants. In addition, intelligence and a number of biological and
neurocognitive parameters were assessed individually (also at school). Teachers were asked to fill out
a brief questionnaire for all TRAILS children in their class. The measures that were used are
described more extensively below.
Subsample with Peer and Teacher Information
We used a subsample of the TRAILS respondents for the analyses. Peer nominations, which
were essential for our study, were only assessed in classes with at least ten TRAILS respondents.
This restriction made the subsample less representative. Children in special education (5.6 percent of
the sample), children in small schools (6.4 percent), and children who repeated (16.9 percent) or
skipped a grade (2.2 percent) were excluded from the subsample. These children did not have many
TRAILS classmates, because our sample is a birth cohort. Children with missing teacher data were
also excluded. The remaining subsample of 918 children (mean age: 11.00, SD = .48; gender: 56.0
percent girls; ethnicity: 8.3 percent children who had at least one parent born in a non-western
country; parent education: 32.5 percent of children had a father and 33.5 a mother with a low
educational level, at maximum a certificate for a lower track of secondary education) differed from
the other TRAILS respondents in several individual and psychosocial characteristics: they were more
often girls, χ2(1, N=2230) = 15.7, p < .01; came on average from higher socio-economic strata,
t(2186)= 5.0, p < .01; had lived more often with the same parents throughout their lives, χ2(1,
N=2230) = 14.1, p < .01; had a higher level of academic performance, t(1923) = 3.2, p < .01; and
were more prosocial, t(1926) = 4.2, p < .01; less aggressive, t(1927) = -3.5, p < .01; and less isolated,
t(1927) = -4.4, p < .01. In sum, the findings can only be generalized to a population of preadolescents
who attend regular elementary schools and did not repeat grades.
Bully-victim dyads The children received a list of all classmates and were asked to nominate
them in a number of dimensions. They nominated their classmates on bullying and victimization,
Bully-Victim Dyads - 11 -
among other things. The number of nominations they could make was unlimited, and the questions
were asked at the dyadic level. Thus, we have bi-directional information on the relations of each pair
of children in a class. Children were not required to nominate anyone. Note that there was no
definition of bullying provided to the children, making it more likely that differences in perspectives
between bully and victim (such as teasing for the perpetrator but bullying for the victim) would come
to the fore.
Our information was based on two peer nomination items: ‘Who do you bully?’ and ‘By whom
are you bullied?’ Children claimed more often to be a victim than a bully, t(917)= 5.3, p < .01. In the
multilevel analyses, our measures can be seen as the aggregates of all the nominations a person gave
to others (as nominator) or received from others (as target), and are for that reason potentially much
more reliable and valid than a self-report (Newcomb, Bukowski, & Pattee, 1993; Cornell &
Brockenbrough, 2004; Salmivalli, 2001).
Nominator and target covariates In close consultation with Masten, we adapted the Revised
Class Play instrument (Masten, Morison, & Pellegrini, 1985). We used the Class Play instrument as a
teacher instead of a peer assessment measure, and we used a five-point answer scale (rating each
child on a range from not applicable to very clearly or frequently applicable). This resulted in reliable
measures. Dominant aggressiveness (Aggressiveness / Disruptiveness) was measured using six items
and had an internal consistency of .89. Sample items were ‘I see the pupil as wanting to be
dominant’, as ‘interrupting others’, or as ‘fighting’. Vulnerability (Isolation / Sensitivity) was
measured using six items and had an internal consistency of .80. Sample items were ‘I see this pupil
as being easily hurt’, as ‘unable to make others listen’, or as ‘having difficulty making friends’.
The number of nominations children received individually from their classmates with regard to
‘best friends’ and ‘dislike’ was used to create measures of peer acceptance and peer rejection, and
these measures were used as individual covariates. The measures were the aggregates of all the
dyadic nominations a person received from others. After the numbers of nominations children
received had been added up, percentages were calculated to take differences in the number of
respondents per class into account, yielding scores from 0 to 1. Sex was also included as a covariate.
Finally, we also took into account whether a child was a bully-victim: 92 of the 918 children
Bully-Victim Dyads - 12 -
belonged to that category. Being a bully-victim was defined as being in the upper quartile for
nominations for bullying as well as victimization (see also Veenstra et al., 2005).
Relationship covariates We took four network characteristics into account and examined
whether bully-victim relationships would be more likely when the nominator disliked the target and
less likely when the nominator liked (i.e., named as a best friend) the target. Furthermore, we
included covariates to measure whether nominator and target had a different sex (mixed-sex) or were
Our data had a three-level structure: networks from 54 school classes (level 3) with 918
individuals (level 2) and 13,606 dyadic relations (level 1). Our dependent variables were dyadic, as
revealed by the questions ‘who bullies who’ and ‘who is victimized by whom’. Our independent
variables were at the individual level (nominator and target covariates), with 918 observations, and at
the dyadic level (relationship covariates), with 13,606 observations. For the analysis of a network
containing binary relationships, we used the p2 model (Van Duijn et al., 2005). The p2 model
incorporates both the individual and the dyadic level, where the different dyads have different
parameters depending on the actors i (‘nominator’) and j (‘target’) involved in the dyad. The p2
model was developed to explain the relationships between actors in a network, using characteristics
of both nominators, targets, and dyads (see also Baerveldt, Van Duijn, Vermeij, & Van Hemert,
2004). The p2 model is an extension of the p1 model (Holland & Leinhardt, 1981). It regards
nominator and target effects not as statistical parameters but as latent (i.e., unobserved) random
variables. These latent variables can be explained by nominator- and target-dependent variables.
Dyad-dependent (relationship covariates) independent variables can also be included in the p2 model.
The p2 model can be regarded as the logistic regression model for dyads and complements the well-
known Social Relations Model (Snijders & Kenny, 1999; Kenny & La Voie, 1984), which is suited
to continuous dyadic outcomes.
In the p2 model, a positive effect of a certain individual or dyadic characteristic can always be
interpreted as having a positive effect on the probability of a relationship. For instance, a positive
Bully-Victim Dyads - 13 -
target effect of sex (where boys are coded as 1, girls as 0) implies that boys have a higher probability
of ‘receiving’, i.e., being reported as a bully by others (either boys or girls). The multilevel version of
the p2 model that we used allowed us to analyze multiple networks simultaneously (Zijlstra et al.,
2006). These p2 analyses are time consuming. The estimation took about 60 hours per model.
Table 1 shows the zero-order correlations for boys and girls. As can be seen, there is an
imperfect match between saying one is a bully and being nominated as a bully (r=.21 for girls and
r=.29 for boys). A similar effect can be observed for saying one is a victim and being nominated as a
victim (r=.31 for girls and r=.20 for girls). This probably indicates that bullying establishes such a
clear-cut relationship that it is consensually identified in most cases. A particular intended act of
bullying may not have been interpreted as such or may have been dwarfed by a more aggressive act.
Conversely, an act of teasing may erroneously have been interpreted as bullying. This makes
correlational analysis less useful for the study of bully-victim matches and underlines the advantages
of analyzing instead bully-victim matches from both points of view and using the multivariate p2
model, including large numbers of relationships from which the matches can be extracted.
Outcomes for ‘Who Bullies Who?’ (Point of view of the bully)
The results of the multilevel p2 model analysis with the question who bullies who as dependent
dyadic variable are given in Table 2. The bully-victim profile hypotheses were that the self-
proclaimed bullies were likely to be (a) boys, (b) dominantly aggressive, and (c) accepted; and they
were matched with victims that were likely to be (d) vulnerable, (e) rejected, and (f) not aggressive.
The data support these hypotheses. We found that being a boy was positively related to being a self-
proclaimed bully in the dyad (b = 1.21, p < .01). A self-proclaimed bully is also likely to be
dominantly aggressive (b = .72, p < .01) and fairly well accepted (b = .23, p = .05). The
characteristics of the victim that is tied to a self-proclaimed bully are also as predicted: vulnerable (b
= .40, p < .01), rejected (b = .30, p < .01), and low in dominant aggressiveness (b = .03, p = .68).
Bully-Victim Dyads - 14 -
The relational hypotheses stated that that the probability of two children having a bully-victim
relationship was (a) larger for boy-boy pairs than for mixed-sex or girl-girl pairs of actors, (b) larger
when the nominator likes the target, and (c) smaller when the nominator dislikes the target. From
Table 2, we see, contrary to our expectation, that the sex composition of the dyad (boy-boy, mixed-
sex, or girl-girl) was unrelated to the probability of a bully-victim relationship occurring. The
relational liking and disliking expectations were supported. When the nominator liked the target, it
was quite unlikely that there would be a bully-victim relationship (b = -.72, p < .01), and when the
target was disliked, the probability of such a relationship was much increased (b = 1.53, p < .01).
Especially dislikes contribute to the likelihood that pupils will interpret their own acts as bullying.
Being a bully-victim was related to being a target (b = 1.75, p < .01) and was not related to
being a nominator in a dyad. We discuss this finding in combination with the results of Table 3.
Without this control variable, we found that aggressiveness was related not only to being a bully, but
also, although much more weakly, to being a victim. This is consistent with the findings of earlier
research by Vermande et al. (2000), who found that a so-called ‘combined central victim/aggressor
model’ was the prevailing pattern of aggressive relationships within school classes.
The overall mean of this model is -6.76; see Table 2. For a dyad with zero scores on all
covariate and random effects, this means that the odds of there being a dyad with a single bully-
victim relation versus a dyad with no ties is .0012. The odds become -6.04 for girls with an
aggressiveness score of one standard deviation above the mean. For aggressive boys (+1 SD), it
becomes -4.83. This results in odds of .0024 for girls and .0080 for boys. Thus, the likelihood of a
dyad occurring with a single bully-victim relation in comparison with a dyad with no ties is very
small, but aggressiveness and being a boy have very large effects on it.
Outcomes for ‘Who Is Victimized by Whom?’ (Point of view of the victim)
The results of the multilevel p2 model analysis with the question who is victimized by whom as
dependent dyadic variable are given in Table 3. The bully-victim profile hypotheses were that self-
proclaimed victims were likely to be (a) vulnerable, (b) rejected, and (c) not aggressive; and they
were matched with bullies who were likely to be (d) boys, (e) dominantly aggressive, and (f) not
Bully-Victim Dyads - 15 -
vulnerable. The data support these hypotheses. As predicted, the self-proclaimed victims are
vulnerable (b = .59, p < .01), rejected (b = .34, p < .01), and low in dominant aggressiveness (b =
.16, p = .09). The bullies to whom the self-proclaimed victims are tied are more likely to be boys (b
= 1.09, p < .01), dominantly aggressive (b = .60, p < .01), and not vulnerable (b = -.35, p <.01). The
bully is fairly well accepted (b = .23, p = .01). Acceptance was predicted not to follow from the goal
of the self-proclaimed victim but to follow from the goal of the self-proclaimed bully. However, the
scores match. This is an extra indication that the results for the two points of view dovetail quite well.
With regard to the relational hypotheses, we see from Table 3 that, again contrary to our
expectation, bully-victim relationships were not more likely in boy-boy relationships and were less
likely in mixed-sex relationships than in girl-girl relationships (b = -.44, p = .04). The relational
liking and disliking expectations were again supported. Liking the target made reporting bullying less
likely (b = -.49, p < .01), and disliking the target made it more likely (b = 2.36, p < .01). Again,
disliking a person seemed to contribute much to the likelihood that pupils would consider themselves
bullied by this person.
As in Table 2, we found that being a bully-victim was related to being a target (b = 1.51, p <
.01) and only marginally related to being a nominator (b = .57, p = .07) in the dyad. Apparently,
being a bully-victim is unrelated to being a self-proclaimed bully or a self-proclaimed victim, but it
contributes considerably to the likelihood that others will report that such a pupil is a bully and a
Lastly, we found in both analyses considerable random actor effects. The nominator variances
were larger than the target variances, indicating that children differ to a large extent in their readiness
to admit or in their perceptions that they are a bully (Table 2) or a victim (Table 3).
Further Examination of the Results
The model fit improved substantially from both perspectives: χ2(16, N=918) = 810.1, p < .01
for who bullies who and χ2(16, N=918) = 1110.1, p < .01 for who is victimized by whom. We also
examined how the well-sampled parameters converged to a stable distribution by inspecting the trace
plots (Zijlstra et al., 2006). All the estimates of the fixed parameters were stable. All random