Childhood Bullying and Social Dilemmas
Amelia Kohm*
Chapin Hall at the University of Chicago, Chicago, Illinois
.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..
Children who witness bullying often do not defend victims. Bystanders might be reticent to intervene because they are stuck in
“social dilemmas.” Social dilemmas are situations in which individuals make decisions based on self-interest due to their lack of
confidence that others will join with them in decisions that benefit the collective. In this study, the social dilemmas concept, which
comes from game theory and social psychology, was applied to bullying for the first time. A total of 292 middle school students at
a private residential school in the United States completed surveys about their bullying-related experiences within their
residences of 10 to 12 students of the same gender. Multilevel modeling was employed to assess if and how attitudes, group norms,
and social dilemmas predict behavior in bullying situations. The findings suggested that both individual and group factors were
associated with behavior in bullying situations and that attitudes, group norms, and social dilemmas each made a unique
contribution to predicting behavior in bullying situations. Aggr. Behav. 41:97–108, 2015. © 2015 Wiley Periodicals, Inc. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..
Keywords: bullying; social dilemmas; bystanders; group norms
INTRODUCTION
Children who witness bullying defend victims in only
12% to 25% of bullying episodes regardless of their
sympathy for the victims or dislike of the bullies (Craig
& Pepler, 2000; O’Connell, 1999). For the bystanders
who decide not to intervene, such bold action may seem
futile at best and dangerous at worst. But might more
bystanders be more willing to intervene if they felt
confident that others would stand with them? As
Aristotle observed long ago: “No tyrant need fear till
men begin to feel confident in each other.”
Evidence from ethnographic studies of children’s
social hierarchies sheds light on the role of bystanders,
who are usually present in bullying situations (Atlas &
Pepler, 1998; Craig & Pepler, 2000; Hawkins, Pepler, &
Craig, 2001;O’Connell, Pepler,&Craig, 1999; Sutton&
Smith, 1999; Xie, Swift, Cairns, & Cairns, 2002). For
example, Adler and Adler (1995); who conducted seven
years of participant-observation and interview research
with third- through sixth-grade students, observed that
most children side with a popular clique member in any
dispute to avoid becoming victimized themselves.
Similarly, a research team that interviewed middle
school and high school students found that harassing and
humiliating weaker, less popular students was a common
method to try to increase their own status at school.
Moreover, victims’ friends rarely defended them and
sometimes joined in the bullying to boost their status
(Bishop et al., 2004).
Such findings, from research on social hierarchies,
provide evidence in line with the hypothesis that social
dilemmas would help to explain why children often do
not defend victims of bullying. Social dilemmas are
situations in which isndividuals make decisions based on
self-interest due to their lack of confidence that others
will join with them in decisions that benefit the collective
(Dawes, McTavish, & Shaklee, 1977; Van Lange,
Liebrand, Messick, &Wilke, 1992). In a common social
dilemma called a public goods dilemma, an individual is
reluctant to contribute to a public good, such as a public
park or clean air, if he or she believes that an insufficient
number of others will also contribute and thus his or her
own efforts would be wasted (Kollock, 1998).
The social dilemmas concept, which comes from
game theory and social psychology, has been applied to
school-age bullying only in the present study, but could
be a fruitful direction for future research. Perhaps
children refrain from defending victims because they
feel that such a selfless contribution (for the good of the
Correspondence to: Amelia Kohm, Chapin Hall at the University of
Chicago, 1313 East 60th Street, Chicago, IL 60637. E-mail: akohm@-
chapinhall.org
Received 29 January 2013; Revised 7 November 2014; Accepted 10
November 2014
DOI: 10.1002/AB.21579
Published online 6 January 2015 in Wiley Online Library
(wileyonlinelibrary.com).
AGGRESSIVE BEHAVIOR
Volume 41, pages 97–108 (2015)
© 2015 Wiley Periodicals, Inc.
victim and the good of the group since it might prevent
future bullying) would be futile unless a sufficient
number of other group members joined their efforts. As
the evidence from ethnographic studies suggests,
unilateral action might leave the defender vulnerable
to victimization (Adler&Adler, 1996; Merten, 1997). In
addition, children might have low expectations about
others supporting a defender because they might
recognize that other children are similarly motivated
to act in their own self-interest.
In a cross-sectional study of 1,220 Finnish elementary
school children, Salmivalli and Voeten (2004) examined
the connections among individual attitudes, group
norms, and students’ roles in bullying situations. Roles
included bullying others, assisting the bully, reinforcing
the bully, defending the victim, or staying outside the
bullying situation, and each role was associated with a
type of behavior. Because the researchers found
behavior in bullying situations to be predicted by not
only individual attributes (attitudes) but also group
characteristics (norms), their study provided an appropriate
model for the current one. In addition, Salmivalli
and Voeten found that when they added attitudes and
norms to their models, there were fairly small reductions
in the variance for behavior in bullying situations, at both
the individual and group levels, suggesting that other
factors are important to predicting behavior in bullying
situations. The current study replicated and extended the
previous study by focusing on whether social dilemmas
help further explain the variance in behavior in bullying
situations.
There were two hypotheses: Both individual factors
(such as attitudes) and group factors (such as norms)
would be associated with behavior in bullying situations;
and attitudes, group norms, and social dilemmas would
each make a unique contribution to predicting student
behavior in bullying situations.
METHOD
Sample and Participant Selection
Participants were 292 middle school students aged 11
to 14 years (29% in sixth grade, 35% in seventh grade,
and 36% in eighth grade; 48.3% were girls) at a private
residential school in the United States that serves
children from low-income families from throughout the
United States. Students at the school are normally
functioning and are not selected according to specific
needs. The racial composition of the school at the time of
the study was approximately 60% Caucasian, 20%
African American, 10% Hispanic, and 10% other.
Students at the school live in residences with other
students of the same gender. A married couple oversees
each residence. At the time of the study, the school had
37 middle school residences each composed of 10 to 12
students in Grades 6, 7, and 8. There was a similar
distribution of students from each grade in each
residence.
Assessments and Measures
Behavior in bullying situations. The Participant
Role Questionnaire (PRQ), developed by Salmivalli
and Voeten (2004), was used in the present study to
assess student behavior in bullying situations within
student residences, the dependent variable. The PRQ
first specifies bullying as when “one child is repeatedly
exposed to harassment and attacks from one or several
other children. Harassment and attacks may be, for
example, shoving or hitting the other one, calling him/
her names or making jokes about him/her, leaving him/
her outside the group, taking his or her things, or any
other behavior meant to hurt another.” The students
reviewed 15 items describing different ways to behave in
such situations and assessed how often each of their
housemates behaved in the ways described since the
school year began (response options are “never,”
“sometimes,” or “often”). The items form five scales
reflecting different participant roles associated with
bullying: bully, assistant, reinforcer, defender, and
outsider. Assistants do not initiate but join in the
bullying; reinforcers encourage the bullying; defenders
help victims; and outsiders are not involved in bullying
in any way (Goldbaum, Craig, & Shelley, 2003; Olthof
& Goossens, 2003; Salmivalli, 1999, 2001; Salmivalli,
Lappalainen, & Lagerspetz, 1998; Sutton & Smith,
1999). The PRQ has demonstrated adequate reliability
and validity in past studies. Cronbach’s alpha coefficients
based on data in the present study were .93 for
the bully scale, .95 for the assistant scale, .93 for the
reinforcer scale, .90 for the defender scale, and .55 for
the outsider scale. Although scores on the bully,
assistant, and reinforcer scales tend to be highly
correlated, according to the authors, they seem to
represent three distinct factors, rather than one underlying
construct (Salmivalli & Voeten, 2004). However,
other studies using the PRQ or an adapted 21-item
version by Sutton and Smith (for younger children)
found that the bully, reinforcer, and assistant roles may
be measuring the same underlying construct (Goldbaum,
Craig, & Shelley, 2003; Sutton & Smith, 1999; Tani,
Greenman, Schneider, & Fregosao, 2003). Thus,
“Composite Pro-bullying Behavior” was also computed
from all items related to bullying, assisting, and
reinforcing behavior.
Attitudes toward bullying. Attitudes toward
bullying were operationalized as students’ moral beliefs
regarding the appropriateness or inappropriateness of
bullying and related behavior (Salmivalli & Voeten,
Aggr. Behav.
98 Amelia Kohm
2004). Students’ attitudes toward bullying were measured
by asking them to evaluate the extent to which they
agreed or disagreed with 10 statements about bullying.
The scale range was .00 to 4.00. Scores were based on
self-reports. Scale means were imputed for missing data
for 23 participants. Higher scores corresponded to more
antibullying attitudes. In the Finnish study, the internal
consistencies of attitudes as measured by Cronbach’s
alpha coefficient was .75. In the present study, the
Cronbach’s alpha coefficient was .73.
Group norms. The development of the questionnaire
designed to assess bullying-related classroom
norms in the Finnish study was guided by the standard
definition of norms as expected standards of behavior in
a certain group (Franzoi, 1996). The norms questionnaire
included questions about behavior that would be
expected or not appropriate in the classroom (Salmivalli
& Voeten, 2004). For each of five situations, students
assessed the likelihood of seven consequences (such as
“Other kids in my residence would avoid him/her” or
“He/she would be considered cool”). A neutral norms
score was also computed based on the sum of the last
item for each condition (“nothing in particular would
take place”). The scale range was 30.00 to 120.00 for the
antibullying scale and 5.00 to 20.00 for the neutral scale.
Higher scores reflected perception of stronger antibullying
norms or neutral norms. Scores were based on selfreport
and aggregated by student residence. The
reliability of the antibullying norm, as measured by
the coefficient alpha, was .90. The alpha for the neutral
norm was .69. In the Finnish study, students were asked
to evaluate the consequences of each act by choosing
from eight optional answers. The present study modified
the norms measure by asking students to evaluate the
probability of several positive and negative consequences
using a Likert scale.
Social dilemmas. A social dilemmas instrument
was developed drawing on the goal-expectation theory
by Pruitt and Kimmel (1977). The theory states that
cooperative behavior arises in a “strategic environment”
(one in which people aim to make rational decisions
toward certain ends) when group members share a goal
of mutual cooperation and an expectation of cooperation
(Pruitt, 1998; Pruitt & Kimmel, 1977). In the present
study, it was assumed (although not measured) that
participants’ decisions regarding bullying were, at least
in part, rational and geared toward certain ends.
Noncooperation or “social dilemmas” arise when
individuals make decisions based on self-interest due
to their lack of confidence that others will join with them
in decisions that benefit the collective. The social
dilemma variable was operationalized as the degree to
which group members agreed that three conditions were
present in their residences: Unilateral action to defend
victims would be dangerous or ineffective; group efforts
could be more effective; and cooperation from others in
an effort to defend a victim was unlikely. Therefore, any
individual’s best short-term strategy was to act selfishly
(i.e., not defend a victim) even though the best long-term
strategy to reduce bullying in the group was to act
collectively (to defend the victim). The instrument
assessed social dilemmas within the context of three
different types of bullying: physical, verbal, and relational.
Eight questions were asked about each type of
bullying. Students were asked to indicate their level of
agreement (strongly agree, agree, disagree, or strongly
disagree) with each item. Examples of items related to
social dilemma conditions for the verbal bullying scale
were, “I could get other kids to stop teasing someone
with other students helping me” and “I could get other
kids to stop teasing someone by myself.” Because
participants were coded as 1 if they met the three social
dilemma conditions and 0 if they did not, means were
equivalent to percentage of participants who met
conditions based on the total number participants who
responded to all relevant questions. If a participant
skipped any of the items related to a condition, that
condition was coded as missing data and it was not
established whether the participant met the criteria for
being in a social dilemma. There was missing data for
7% to 23% of the participants, depending on the type of
bullying under consideration. Means were not imputed
for missing data because conditions were not established
based on scales composed of similar items. Therefore,
there were no logical means to impute. The reliability of
the six items related to social dilemma conditions was
then assessed for each type of bullying. The alpha for
verbal bullying (teasing) was .64, for physical bullying
(beating up or pushing around) was .71, and for
relational bullying (gossiping) was .67. The reliability
of all 18 items, assessed together, was .87.
Minor edits were made to the original items used in the
Finnish study to make them more understandable to
students at the participating school. Edits were based on
feedback during pilot testing of the instrument with 15
students at the school, in Grades 4 to 8.
Assessment of missing data. Missing data
points were replaced with imputed scale means or
sample means for categorical variables. Mean substitution
produces internally consistent sets of results.
However, it also artificially decreases the variation of
scores, and this decrease is proportional to the amount of
missing data. To assess the possible effect of missing
data, dummy variables were created as controls where
means were imputed for missing data. Only a small
number of the coefficients for the dummy variables for
the missing data were significant, suggesting that those
participants with missing data did not significantly differ
Aggr. Behav.
Childhood Bullying and Social Dilemmas 99
from thosewithout missing data. In addition, final models
for three outcomes (composite probullying, withdrawing,
and defending) were run with cases with missing data
deleted. This procedure eliminated approximately one
third of the cases (n¼190). However, even with this
much-reduced sample, the coefficients were generally
similar in size and direction to those produced with the
whole sample (which included imputed means and
dummy variable controls) suggesting that the missing
data did not have a substantial effect on the results.
Procedure
Parents and guardians of all students in the designated
grades (N¼389) were contacted to inform them of the
study, explain their children’s rights as participants, and
ask if they would like their child to participate. The
transience of this low-income population often made it
difficult to reach parents and guardians. Consent was
received from 308 (or 79%) of the parents and guardians
and refusal from 19 (or 5%). A total of 292 students
living in 37 student residence (or 95% of the students
who had parental consent) agreed to complete the
questionnaire.
The data were collected via online questionnaires in
the computer laboratory at the middle school building.
Those students with parental consent who also assented
to participate in the study completed the questionnaire
and were led through the four instruments in the
following order: attitudes measure, PRQ, social dilemmas
measure, and norms measure.
Analyses
To assess if and how attitudes, group norms, and social
dilemmas predict behavior in bullying situations,
multilevel modeling was employed using Hierarchical
Linear Modeling software (Raudenbush & Bryk, 2002).
Multilevel modeling is a type of regression analysis
designed to handle hierarchical or clustered data. In the
current study, students were considered Level 1 units and
were clustered in residences that were considered Level
2 units. Observations of students within groups were
likely to be more similar than observations of individual
students sampled from different residences. When such
conditions exist, there is an intraclass correlation (ICC),
and the assumption of independence of observations for
regular regression is violated (Hox, 1998; Kreft & De
Leeuw, 1998; Raudenbush & Bryk, 2002). In the current
study, the groups of interest were the residences, rather
than classrooms, because ICCs are typically higher in
family households than in classrooms, and residences
within an educational setting might be more similar to
households (Gulliford, Ukoumunne, & Chinn, 1999;
Murray et al., 1994; Siddiqui, Hedeker, Flay, & Hu,
1996).
A series of 11 multilevel regression models of
increasing complexity were run for each of the dependent
variables: probullying (a composite of the three
probullying behavior), withdrawing from bullying
situations, and defending victims of bullying.1 Each
series of regressions began with a null model, which
included an intercept and two variance components:
behavior differences between students within residences
and behavior differences between residence behavior
means. The null model served as a reference for
subsequent models, each of which included variables
from previous models and an additional variable of
interest.
Variables that controlled for missing data and/or
significant interactions between key variables with
gender or grade were added, along with variables of
interest as appropriate. Interactions between grade and
gender with nonsignificant coefficients were not included
in models. Grade and attitudes were entered into
the model as Level 1 predictors of behavior in bullying
situations. Such predictors could explain both withinand
between-group variances because each residence
had a different group of students. Gender and norms
were entered into the model as Level 2 predictors.
Because there was only one gender per residence, this
variable could not explain within-group variance.
Similarly, because the group norms variables were
aggregated to the group level, they could only explain
variance between groups. The social dilemma variable
was entered as a Level 1 predictor (i.e., whether the
individual reported all three social dilemma conditions)
and, in aggregate form, as a Level 2 predictor (i.e., the
percent age of residence members who reported all three
social dilemma conditions). The attitudes and norms
predictors were continuous variables whereas the grade,
gender, and social dilemma variables were dummy
variables. Unlike the study by Salmivalli and Voeten
(2004), the current study did not omit the general
intercept and thereby create separate coefficient estimates
for each grade. To simplify analyses, Grade 6 was
used as the reference category for grade.
As in the study by Salmivalli and Voeten (2004) a
Rankit transformation was employed to reduce the
influence of outliers and normalize the distribution of
behavioral variables (Noruésis, 1993). However, the
distribution of the raw scores did not strongly depart
from normality as it did in the earlier study. In addition,
for each model that added a Level 1 variable, the model
was run twice: once with the slope fixed (or set to 0) at
Level 2 and once with a random slope, one that is
1 This article does not report results on models for the individual
probullying roles-bully, reinforcer, and assistant-due to evidence that they
may be measuring the same underlying construct.
Aggr. Behav.
100 Amelia Kohm
allowed to vary across groups. The deviance statistics for
the two models were then compared with each other,
taking into account the number of parameters in each
model, using a chi-square test. In none of the tests was a
difference statistically significant. Thus all slopes for
Level 1 variables were fixed at Level 2, meaning that the
relation between individual-level variables and behavior
outcomes did not vary by residence.
RESULTS
Descriptive
Table I presents the means and standard deviations
of boys and girls in the three grade levels for each of
the variables2. The means and standard deviations
for the antibullying and neutral norms were residence
averages and were not categorized by grade
because each residence included students in all three
grades.
Reinforcing bullies, defending victims, or withdrawing
from bullying situations were more common
behavior than bullying and assisting a bully. Assisting
the bully seemed to decrease with age for girls in the
sample. Also, there was an increasing trend, from sixth
to eighth grade, in both defending victims and withdrawing
in bullying situations for both boys and girls.
Bullying, assisting, reinforcing, and withdrawing were
more prevalent among boys than girls, while defending
was more prevalent among girls. With respect to
attitudes, girls’ antibullying attitudes appeared to
decrease with age. Boys and girls did not appear to
differ in the strength of their antibullying attitudes. In
addition, boys’ and girls’ residences were similar in the
strength of antibullying and neutral norms.
The number of students who reported all three
conditions for social dilemmas with regard to verbal
bullying seemed to decrease with age for boys and
increase with age for girls. The number of girls reporting
social dilemma conditions related to relational bullying
also appeared to increase with age. In addition, the
number of girls reporting social dilemma conditions
related to physical bullying seemed to decrease with age.
The majority of students had witnessed bullying in
their residences, with verbal (96.2%) and relational
(91.8%) being the most common types of bullying. In
addition, 50% to 60% of the students, depending on the
type of bullying, believed that unilateral efforts to help
victims would be dangerous and/or ineffective in their
residences. Similarly, 44% to 54% believed that group
efforts would be more effective and/or safe. Fewer
students (approximately 30% for each type of bullying)
had low expectations that their housemates would help
them defend a victim.
Assessment of Regression Coefficients
Table II provides the regression coefficients for the
variables of interest and related standard errors for the
final model for each of the dependent variables. Because
coefficients were not standardized, comparisons across
predictors should be considered relative to their standard
errors. Review of the coefficients begins with withingroup
(Level 1) predictors and then focuses on betweengroup
(Level 2) predictors.
Grade was entered as a dummy variable with Grade 6
as the reference category. The differences between
seventh graders’ and sixth graders’ behavior in bullying
situations were not statistically significant. Similarly,
eighth graders did not differ from sixth graders in terms
of their behavior, except with regard to defending
victims. Students in the eighth grade, on average, ranked
significantly higher on defending behavior than those in
the sixth grade.
Consistent with Hypothesis 2, the coefficients suggested
that as antibullying attitudes increased, probullying
ranks decreased and defending ranks increased.
Antibullying attitudes, however, did not have a
significant effect on withdrawing behavior.
Contrary to expectations, within-group variation on
reporting social dilemma conditions generally did not
predict behavior in bullying situations. All coefficients
were nonsignificant, with the exception of those for
social dilemmas related to relational bullying. The
coefficient for this variable was significant in the model
for the probullying composite outcome. There was an
inverse relation between reporting social dilemma
conditions and probullying behavior. Thus, contrary to
expectations, those who reported all three conditions
tended to rank lower on probullying behavior. However,
the effect size, given that the values for social dilemmas
could only be 1 or 0, was modest.
The coefficient on gender was significant for all
behavior except defending. In general, boys’ residences
ranked higher than girls’ residences on probullying
behavior and on withdrawing behavior.
The relation of antibullying norms to probullying
behavior was significant and in the expected direction:
As antibullying norms increased, probullying behavior
ranks decreased. The coefficient for withdrawing
behavior was not significant, and the antibullying norms
coefficient for defending behavior approached significance
and was positive, as expected. Student residences
that ranked higher on neutral norms tended to have
students who ranked higher on probullying behavior and
2 Description in this section is based on inspection of the descriptive data in
Table 1 and no inferential statistics were carried out.
Aggr. Behav.
Childhood Bullying and Social Dilemmas 101
lower on defending, although the coefficient for the
composite probullying outcome only approached significance.
In addition, the coefficient for neutral norms
related to withdrawing behavior was not significant.
The number of students in a residence reporting all
three social dilemma conditions related to either
physical or relational bullying tended to have a positive
relation with probullying behavior and withdrawing, as
expected. None of the coefficients for mean social
dilemmas related to verbal bullying were significant
(however, note the interactions discussed below).
In addition, several significant interactions indicated
that the relation between mean social dilemmas and
behavior sometimes varied by gender or grade. As shown
in Figure 1, mean social dilemmas related to physical
bullying did not predict withdrawing behavior for boys,
but did predict this behavior for girls: on average, girls
scored far below the mean on withdrawing behavior in
residences with low mean scores (mean1 SD) and
scored slightly above the mean on withdrawing behavior
in residences with high mean scores (meanþ1 SD). In
addition, mean social dilemmas related to verbal bullying
did not predict girls’ defending behavior, but they
appeared to be related to boys’ defending behavior.
Boys in residences with low mean social dilemmas
related to verbal bullying (mean1 SD) tended to score
above the mean on defending behavior whereas boys in
residences with high social dilemmas (meanþ1 SD)
tended to score below the mean on defending behavior. A
significant interaction between grade and mean social
dilemma related to relational bullying was detected in the
final defender model. Sixth graders’ defending behavior,
on average, was not strongly associated with the number
of housemates reporting social dilemma conditions
related to relational bullying. However, eighth-grade
students in residences with low mean social dilemmas
related to relational bullying (mean1 SD) tended to
rank substantially higher on defending behavior than
those in high mean social dilemma residences (meanþ1
SD). However, it should be noted that even students in
residences with high mean social dilemmas tended to
score above the mean on defending behavior.
Assessment of Variance Components
In addition to coefficient statistics, Hierarchical
Linear Modeling also results in data on the variance
components of each model. Variance is analogous to
the error term in traditional regression equations. The
multilevel model disaggregated the total variation into
a component at the individual level (i.e., withinresidence
variation) and at the group level (i.e.,
between-residence variation).
TABLE I. Score Means (and Standard Deviations) of Boys and Girls From Different Grade Levels
Independent Variable Gender Grade 6 Grade 7 Grade 8
Behaviors
Bullying Boys .59 (.35) .70 (.38) .65 (.36)
Girls .54 (.40) .56 (.36) .48 (.34)
Assisting the bully Boys .66 (.36) .74 (.37) .71 (.35)
Girls .64 (.42) .62 (.39) .58 (.32)
Reinforcing the bully Boys .86 (.35) .92 (.35) .88 (.35)
Girls .87 (.41) .81 (.36) .80 (.37)
Defending the victim Boys .79 (.29) .84 (.31) .96 (.33)
Girls .83 (.26) .86 (.29) .94 (.31)
Withdrawing Boys .92 (.20) .93 (.18) .96 (.18)
Girls .88 (.19) .87 (.18) .91 (.13)
Attitudes
Boys 3.17 (.55) 2.75 (.53) 2.78 (.60)
Girls 3.02 (.64) 2.95 (.51) 2.80 (.54)
Group Norms
Antibullying norms Boys’ Homes 69.35 (11.45)
Girls’ Homes 69.45 (14.24)
Neutral norms Boys’ Homes 12.88 (3.08)
Girls’ Homes 12.77 (2.84)
Social Dilemmas
Verbal Bullying: Meets Conditions A, B, C Boys .19 (.39) .18 (.39) .10 (.31)
Girls .15 (.36) .17 (.38) .23 (.42)
Physical Bullying: Meets Conditions A, B, C Boys .10 (.30) .14 (.35) .10 (.31)
Girls .22 (.42) .16 (.37) .10 (.30)
Relational Bullying: Meets Conditions A, B, C Boys .07 (.26) .21 (.41) .04 (.20)
Girls .08 (.28) .11 (.32) .14 (.35)
Aggr. Behav.
102 Amelia Kohm
As indicated in Table II, ICCs for outcome behavior in
the present study ranged from 11.5% for defending and
withdrawing behavior to 19% to 20% for probullying
behavior. This finding supported Hypothesis 1 that both
individual and group factors would be associated with
behavior in bullying situations. In addition, although
there were clear associations between context and
behavior for all of the behavior measured, the probullying
behavior was more closely associated with context
than were withdrawing and defending.
Because none of the models included random slopes, it
was possible to compute explained (or modeled)
variance, analogous to R2 statistics in traditional
regression. These figures were computed by subtracting
the variances of the present model from the variances of
the null model and dividing by the variances of the null
model. Thus, they showed the proportion of total
variance at each level that was explained after the
addition of the variable to the present model. In some
cases, adding predictors to a model actually increased
the variance and thus decreased the variance explained.
These predictors unnecessarily complicated the models,
using up degrees of freedom and thus increasing
variance. In addition, if a predictor that models part of
the within-group variability does not model part of the
between-group variability, the decrease in the Level 1
variance must be balanced by an increase in the estimate
of the Level 2 variance. Adding a Level 1 predictor
results in a decrease in the similarity within groups and,
consequently, an increase in the dissimilarity between
groups (Snijders & Bosker, 1994).
As shown in Table II, the variable that explained the
most within-group variance for the probullying behavior
outcome was antibullying attitudes. None of the other
predictors resulted in sizeable decreases in variances.
Indeed, while the ICCs indicated that the majority of the
TABLE II. Total Explained Variance (R2) for Each Model
Behavior and ICC Model Variables Added (Level Added) Level 1 Variance Level 2 Variance R2W (%) R2B (%)
Composite Probully Behavior Null .815 .202
1 Grade .817 .203 0.25 0.50
ICC: 19.86% 2 Gender .817 .185 0.25 8.42
3 Antibullying Attitudes .732 .157 10.18 22.28
4 Antibullying Norms .73 .145 10.43 28.22
5 Neutral Norms .73 .127 10.43 37.13
6 Social Dilemma – Verbal Bullying .725 .137 11.04 32.18
7 Social Dilemma – Physical Bullying .731 .135 10.31 33.17
8 Social Dilemma – Relational Bullying .723 .144 11.29 28.71
9 Social Dilemma – Verbal Bullying .723 .138 11.29 31.68
10 Social Dilemma – Physical Bullying .723 .101 11.29 50.00
11 Social Dilemma – Relational Bullying .725 .077 11.04 61.88
Withdrawing Null .888 .116
1 Grade .876 .127 1.35 9.48
ICC: 11.55% 2 Gender .874 .114 1.58 1.72
3 Antibullying Attitudes .875 .117 1.46 0.86
4 Antibullying Norms .875 .129 1.46 11.21
5 Neutral Norms .875 .123 1.46 6.03
6 Social Dilemma – Verbal Bullying .877 .121 1.24 4.31
7 Social Dilemma – Physical Bullying .87 .124 2.03 6.90
8 Social Dilemma – Relational Bullying .873 .126 1.69 8.62
9 Social Dilemma – Verbal Bullying .873 .138 1.69 18.97
10 Social Dilemma – Physical Bullying .872 .107 1.80 7.76
11 Social Dilemma – Relational Bullying .873 .078 1.69 32.76
Defending the Victim Null .89 .116
1 Grade .851 .126 4.38 8.62
ICC: 11.53% 2 Gender .851 .133 4.38 14.66
3 Antibullying Attitudes .826 .114 7.19 1.72
4 Antibullying Norms .825 .095 7.30 18.10
5 Neutral Norms .824 .092 7.42 20.69
6 Social Dilemma – Verbal Bullying .829 .084 6.85 27.59
7 Social Dilemma – Physical Bullying .833 .076 6.40 34.48
8 Social Dilemma – Relational Bullying .836 .079 6.07 31.90
9 Social Dilemma – Verbal Bullying .84 .019 5.62 83.62
10 Social Dilemma – Physical Bullying .838 .023 5.84 80.17
11 Social Dilemma – Relational Bullying .821 .003 7.75 97.41
Note. R2W, within-group variance; R2B, between-group variance.
Aggr. Behav.
Childhood Bullying and Social Dilemmas 103
variance in the behavior is explained at the individual
level, the predictors included in the present study’s
models did not explain much of that variance. For the
probullying behavior model, the predictors explained
between 10% and 13% of the Level 1 variance. Eight
percent of the variance for defending behavior and only
2% of the variance for withdrawing were explained by
the predictors in the models. The predictors accounted
for significantly more of the Level 2 variance. For the
probullying models, the predictors explained 54% to
62% of the Level 2 variance. The predictors accounted
for 33% of the between-group variance in withdrawing
behavior and 97% of the between-group variance in
defending behavior.
For the probullying models (with the exception of
reinforcing), adding gender resulted in sizeable increases
in explained variance, particularly for bullying
behavior (from 1.46% to 17.01%). The addition of
antibullying attitudes resulted in even larger increases in
explained variance. (Although attitudes were added at
the individual level, housemates’ similarity in attitudes
resulted in reductions in Level 2 variance.) For example,
in the composite probullying model, adding antibullying
attitudes increased explained variance from 8.42% to
22.28%. Antibullying norms and neutral norms also
generally resulted in sizeable reductions in Level 2
variance for probullying behavior. Finally, although
adding the individual reports of social dilemma
conditions had very little effect on the overall explained
variance in the probullying models, adding the mean
social dilemma variables related to physical and relational
bullying resulted in sizeable increases in
explained Level 2 variance. For example, in the
composite probullying model, adding mean social
dilemmas related to physical bullying increased explained
variance from 31.68% to 50.00%.
For the withdrawing models, most of the predictors
added resulted in decreases in explained Level 2
variance. Indeed, the only predictors that had a
substantial effect were the mean social dilemma
variables related to physical and relational bullying.
Adding the mean social dilemma variable related to
relational bullying increased explained variance at Level
2 from 7.76% to 32.76%.
For the defending models, the additions of antibullying
norms, mean social dilemmas related to verbal
bullying, and mean social dilemmas related to relational
bullying each resulted in substantial increases in
explained variance. For example, adding mean social
dilemmas related to verbal bullying increased explained
variance from 31.90% to 83.62%.
DISCUSSION
The present study produced support for both hypotheses:
(1) Both group and individual factors predicted
behavior in bullying situations; and (2) Attitudes, group
norms, and social dilemmas each made a unique
contribution to predicting student behavior in bullying
situations. For the outcome behavior in the present study,
-0.90
-0.80
-0.70
-0.60
-0.50
-0.40
-0.30
-0.20
-0.10
0.00
0.10
0.20
Low Mean High
Withdrawing Behavior Z-Score
Mean Social Dilemma Related to Physical Bullying
Girls
Boys
Fig.1. Expected ranks for withdrawing behavior X mean social dilemmas (physical) X gender.
Aggr. Behav.
104 Amelia Kohm
the ICCs ranged from 11.5% for defending and
withdrawing behavior to between 19% and 20% for
the probullying behavior. These findings supported the
hypothesis that both individual and group factors would
be associated with behavior in bullying situations.
Moreover, the probullying behaviors were more closely
associated with group factors than were withdrawing
and defending. Salmivalli and Voeten (2004) did not
report on ICCs for their models, although they detected
variance at both levels, suggesting that both individual
and group factors predicted behavior in bullying
situations.
As predicted by Hypothesis 2, results indicated that
antibullying attitudes were predictors of behavior in
bullying situations and that as antibullying attitudes
increased, probullying ranks decreased and defending
ranks increased. Antibullying attitudes did not appear to
have a significant association with withdrawing behavior.
Salmivalli and Voeten (2004) also found that
antibullying attitudes were inversely related to probullying
behavior and positively related to defending
behavior. However, unlike the present study, Salmivalli
and Voeten also found that antibullying attitudes were
positively related to withdrawing behavior, which is in
line with the hypotheses for the present study.
The two samples might have had different societal
norms concerning withdrawing behavior (which are not
measured in either study). Students in the Finnish
sample, who opposed bullying, might have felt that
withdrawing is an acceptable response in bullying
situations, whereas students with antibullying attitudes
in the present study might have felt that defending is a
more acceptable response. Investigation into the differences
in cultural norms related to antisocial behavior,
behavior in bullying situations, and social behavior in
general between the United States and Finland and how
these differences vary by age, race, region, institution,
and gender fell outside of the scope of this study.
Although some research has been conducted comparing
the prevalence of bullying between countries, there is a
dearth of research comparing attitudes toward bullying,
their relation to behavior in bullying situations, and
possible reasons (such as cultural norms) for differences
in behavior between countries (Nansel, Craig, Overpeck,
Saluja,&Ruan, 2004). The results of one study indicated
few differences in England and Italy in children’s
attitudes toward bullying (Menesini et al., 1997).
Another study compared moral emotions and reasoning
to children’s behavior in bullying situations in Spain and
Italy. The research team found differences in egocentric
disengagement motives between Italian and Spanish
students and speculated on cultural norms that might
account for such differences (Menesini et al., 2003). No
similar comparisons between American and Finnish
students have been conducted. Other research has shown
a link between attitudes and behavior; although people
strive for attitude-behavior consistency, it is not always
clear whether attitudes cause behavior or vice versa
(Eagly & Chaiken, 1993).
The present study showed that antibullying and neutral
norms were group factors associated with probullying
behavior and defending behavior. Specifically, and
consistent with Hypothesis 2, as antibullying norms
increased, probullying behavior ranks decreased. The
antibullying norms’ coefficient for defending behavior
approached significance and was positive, as expected.
Student residences that rank higher on neutral norms
tended to have students who ranked higher on probullying
behavior and lower on defending. In addition, norms did
not appear to have a significant association with withdrawing
behavior. There were no significant interactions
between either norms variable with gender or grade in the
present study. Similarly, Salmivalli and Voeten (2004)
found that antibullying norms were negatively associated
with bullying and reinforcing behavior—but only for
fifth- and sixth-grade students—and positively associated
with defending behavior—but only for sixth-grade
students. Fourth-grade students’ behavior was generally
not associated with antibullying norms, although these
students were more likely to withdraw when antibullying
norms were lower than were students in the other grades.
Neutral norms had variable relations with behavior in the
Finnish study, depending on the grade and gender of
participants. In the present study, by contrast, more
positive relations were found between neutral norms and
probullying behavior. There was no significant relation
between neutral norms and withdrawing behavior in
either study. In addition, in both studies, there was a
negative relation between neutral norms and defending
behavior.
Because students in the present study attended a
residential school that promotes a certain school identity,
they might have been more influenced by school-wide
norms than were the Finnish children who attended day
schools, who, by contrast, might have been more
influenced by their immediate classmates. However,
because we measured norms in a somewhat different
manner in the present study than in the Finnish study,
differences in the findings should not be over interpreted.
In line with the current study, a number of studies
suggest that children tend to behave in ways that are
deemed acceptable by others in their particular group,
and behavior related to aggression and social withdrawal
appears to be particularly influenced by classroom
norms, while prosocial behavior does not appear to be as
closely linked with norms (Chang, 2004; Stormshak
et al., 1999Stormshak, Bierman, Bruschi, Dodge, &
Coie, 1999).
Aggr. Behav.
Childhood Bullying and Social Dilemmas 105
The findings of the present study suggested that social
dilemma dynamics help predict behavior in bullying
situations. Moreover, unlike attitudes and norms, social
dilemmas helped predict withdrawing behavior. In
addition, and contrary to expectations, within-group
variation in reporting social dilemma conditions generally
did not predict behavior in bullying situations.
However, as expected, the number of students in a
residence reporting all three social dilemma conditions
related to either physical or relational bullying tended to
have a positive relation with probullying behavior and
withdrawing. It should be noted that most of the
coefficients for mean social dilemmas related to verbal
bullying were not significant (with the exception of
boys’ defending behavior). In addition, several significant
interactions indicated that the relation between
mean social dilemmas and behavior sometimes varied
by gender or grade. Salmivalli and Voeten (2004) did not
measure social dilemmas in their study, nor have any
other studies examined the relation between social
dilemmas and behavior in bullying situations. However,
the findings regarding social dilemmas in the present
study raised some important questions, primarily: (1)
Why did social dilemmas predict behavior only at the
group level? and (2) Why were social dilemmas better at
predicting withdrawing behavior than were attitudes and
norms? These issues are addressed below.
Why Did Social Dilemmas Predict Only at the
Group Level?
Goal-Expectation Theory predicts that an individual,
under social dilemma conditions, looks at the situation
and understands that he or she is contributing to the
problem but believes that a unilateral effort will have no
impact on the situation; only a group effort will work.
Moreover, because he or she has low expectations that
enough other people will act in the interest of the group,
he or she concludes that it is pointless to act in the
interest of the group. In the present study, the social
dilemma predictors at the individual level (whether a
student agreed that the three social dilemma conditions
existed in his or her residence) did not predict behavior
well. However, students in residences where more
students reported social dilemma conditions—regardless
of their own assessment of social dilemma
conditions—scored higher on probullying behavior
and withdrawing and lower on defending behavior.
One possible interpretation is that the more students who
reported social dilemma conditions, the more likely it
was that those conditions actually existed. To date, the
most common method used in social dilemma research
has been laboratory experiments in which researchers
develop “games” that include social dilemma conditions
and then observe how participants behave in those
situations (Johnson & Johnson, 2001; Pellegrini, 2002;
Piliavin, 2001). In addition, the relatively few field
studies that have been conducted usually started with a
situation in which social dilemma conditions naturally
exist and then asked respondents how they behaved and
why (Fujii, Garling, & Kitamura, 2001; Ohnuma,
Hirose, Karasawa, Yorifuji, & Sugiura, 2005; Tyler &
Degoey, 1995). The present study relied on students’
perceptions to establish whether social dilemma conditions
existed within various residences. Thus it is
possible that, even though an individual within a
residence did not perceive the conditions, he or she
was in a residence that had the conditions, and was acting
accordingly, albeit not consciously. Another interpretation
might be that individuals’ low expectations of peers
led them to conclude that such behavior is the norm (a
norm not measured by the instrument used to measure
norms in the study since both the norms instrument and
the social dilemmas instrument explained unique
variance). The perceived norm, in turn, led them to
behave “noncooperatively” and perhaps to adjust their
attitudes accordingly.
Why Were Social Dilemmas Better at
Predicting Withdrawing Behavior Than
Were Attitudes and Norms?
Residences with more students perceiving social
dilemma conditions were characterized, in particular,
by a larger number of students reporting that they did not
expect their housemates to defend a victim in a bullying
situation. If students in such residences expected their
peers to withdraw from bullying rather than defend a
victim, then these students might have been more likely
to act in kind to conform to a withdrawing norm.
Limitations
Although the study produced interesting findings that
warrant further investigation, it is important to note its
limitations. The cross-sectional design did not allow an
assessment of the causal direction between the predictors
and behavior. Thus, although one hypothesis was
that attitudes, norms, and social dilemmas would lead to
certain behavior in bullying situations, it could be that
the behavior led to the attitudes, norms, and/or social
dilemmas. For example, sometimes individuals infer
their attitudes from their behavior. In addition, the
statistically significant associations that were found
between the predictor variables and the outcome
variables could result from both predictors, in a
statistical sense, and outcomes being associated with a
third, unmeasured, variable.
Missing data might also limit the reliability of the
findings of the present study. The response rate was 75%
of the total middle school population. The lack of
Aggr. Behav.
106 Amelia Kohm
participation by 25% of the population primarily was
due to parents and guardians not responding to the
request for consent. In addition, 34.9% of participants
had missing data on at least one predictor, and 10 (of 37)
residences had 50% or more students (who participated
in the study) with missing data on at least one predictor
variable. As discussed above, assessment of the impact
of missing data suggested that those participants with
missing data did not significantly differ from those
without missing data.
The instruments employed to measure the variables of
interest also had potential limitations. The PRQ
generally shows good psychometric properties, and
the bully, assistant, and reinforcer roles appear conceptually
distinct. However, the subscales used to assess
these three probullying roles might be measuring the
same underlying concept, according to results from
earlier studies. Thus, a conservative approach to
interpretation of findings would focus only on the
“composite probullying role.” In addition, as noted, the
outsider scale’s reliability was low in the present study
(the Cronbach’s alpha coefficient was .55), which
limited understanding of how the independent variables
were related to this dependent variable.
Another limitation might have been the way social
dilemmas were measured. As noted above, the study
relied on students’ perception of the conditions of a
social dilemma because there was no way to clearly
establish the existence of those conditions as one might
in situations in which the costs and benefits of acting
selfishly and cooperatively can be objectively demonstrated
as in laboratory studies. However, relying on
perceptions may be problematic with middle school
students, who might not be socially sophisticated enough
to understand the costs and benefits of unilateral versus
multilateral action. Social dilemmas also might have
been measured more accurately had more items related
to each social dilemma condition been included in the
survey instrument. In the present study, there were only
one to two items per condition, which did not allow a
very rigorous testing of reliability. Moreover, social
dilemmas were treated as a categorical variable (i.e.,
social dilemma conditions either existed or did not exist,
according to student reports). A continuous variable
might have provided a more subtle understanding of how
such conditions, as they grow stronger, affect individual
and group behavior.
Contribution
The study furthered inquiry into group factors related
to bullying. Past research in this area has focused on
norms as the key group factor that might affect bullying.
The current study examined another group factor: the
role of social dilemmas in bullying. The study also
contributed to the literature on real-life social dilemmas.
Social dilemma research has been criticized for relying
on computer simulations and laboratory experiments, in
which real or virtual participants play games that present
dilemmas. If future research supports the importance of
unraveling social dilemmas to the reduction of bullying
in schools, then interventions that employ strategies that
tend to moderate social dilemmas in other types of
circumstances might be tested.
Bullying research, in recent years, has focused more
on the role of bystanders in encouraging bullying or
passively allowing it to continue. The present study
provides a possible explanation for their behavior and a
possible direction for future interventions.
ACKNOWLEDGMENTS
I am unable to reveal the funding source of the
research because I must maintain the confidentiality
assurances made to the University of Chicago Human
Subjects Committee to protect identifying information
of the human participants in the research. There is also
the same requirement to protect human participants’
identifying information in the agreement with the funder
as a condition of the award.
REFERENCES
Adler, P. A., & Adler, P. (1995). Dynamics of inclusion and exclusion in
preadolescent cliques. Social Psychology Quarterly, 58, 145–162. doi:
10.2307/2787039.
Adler, P. A., & Adler, P. (1996). Preadolescent clique stratification and the
hierarchy of identity. Sociological Inquiry, 66, 111–142. doi: 10.1111/
j.1475–682x.1996.tb00213.x.
Atlas, R. S., & Pepler, D. J. (1998). Observations of bullying in the
classroom. Journal of Educational Research, 92, 86–99. doi: 10.1080/
00220679809597580.
Bishop, J. H., Bishop, M., Bishop, M., Gelbwasser, L., Green, S., Peterson,
E., & Zuckerman, A. (2004). Why we harass nerds and freaks: A
formal theory of student culture and norms. Journal of School Health,
74, 235–251. doi: 10.1111/j.1746–1561.2004.tb08280.x.
Chang, L. (2004). The role of classroom norms in contextualizing the
relations of children’s social behavior to peer acceptance. Developmental
Psychology, 40, 691–702. doi: 10.1037/0012–
1649.40.5.691.
Craig, W. M., & Pepler, D. J. (2000). Observations of bullying and
victimization in the school yard. InW.M. Craig (Ed.), Childhood social
development: The essential readings, (pp. 116–136). Malden, MA:
Blackwell Publishers.
Dawes, R. M., McTavish, J., & Shaklee, H. (1977). Behavior,
communication, and assumptions about other people’s behavior in a
commons dilemma situation. Journal of Personality and Social
Psychology, 35, 1–11. doi: 10.1037/0022–3514.35.1.1.
Eagly, A. H., & Chaiken, S. (1993). The psychology of attitudes, Fort
Worth, TX: Harcourt Brace Jovanovich College Publishers.
Franzoi, S. (1996). Social Psychology, Dubuque, IA: Time Mirror Higher
Education Group, Inc.
Fujii, S., Garling, T., & Kitamura, R. (2001). Changes in drivers’
perceptions and use of public transport during a freeway closure:
Effects of temporary structural change on cooperation in a real-life
Aggr. Behav.
Childhood Bullying and Social Dilemmas 107
social dilemma. Environment and Behavior, 33, 796–808. doi:
10.1177/00139160121973241.
Goldbaum, S., Craig,W. M., & Shelley, D. (2003). Predicting participant
roles in bullying. Paper session presented at the Society for Research in
Child Development Biennial Meeting, Tampa, FL.
Gulliford, M. C., Ukoumunne, O. C., & Chinn, S. (1999). Components of
variance and intraclass correlations for the design of community-based
surveys and intervention studies: Data from the health survey for
England 1994. American Journal of Epidemiology, 149, 876–883. doi:
10.1093/oxfordjournals.aje.a009904.
Hawkins, D. L., Pepler, D. J., & Craig, W. M. (2001). Naturalistic
observations of peer interventions in bullying. Social Development,
10, 512–527. doi: 10.1111/1467–9507.00178.
Hox, J. (1998). Multilevel modeling: When and why. In I. Balderjahn R.
Mathar & M. Schader (Eds.), Classification, data analysis, and data
highways: Proceedings of the 21st Annual Conference of the
Gesellschaft für Klassifikation e.V., University of Potsdam, March 12–
14, 1997, (pp. ix, 414) New York: Springer.
Johnson, D. W., & Johnson, R. T. (2001). Psychology of cooperation and
competition. In N. J. Smelser & P. B. Baltes (Eds.), International
encyclopedia of the social and behavioral sciences, Oxford, UK:
Elsevier Science.
Kollock, P. (1998). Social dilemmas: The anatomy of cooperation. Annual
Review of Sociology, 24, 183–214. doi: 10.1146/annurev.soc.24.1.183.
Kreft, I., & De Leeuw, J. (1998). Introducing multilevel modeling.
London: Sage.
Menesini, E., Eslea, M., Smith, P. K., Genta, M. L., Giannetti, E., Fonzi,
A., & Costabile, A. (1997). Cross-national comparison of children’s
attitudes towards bully/victim problems in school. Aggressive
Behavior, 23, 245–257. doi: 10.1002/(sici)1098–2337(1997)
23:4 (245::aid-ab3 ) 3.0.co;2-j.
Menesini, E., Sanchez, V., Fonzi, A., Ortega, R., Costabile, A., & Lo, F.
(2003). Moral emotions and bullying: A cross-national comparison of
differences between bullies, victims and outsiders. Aggressive
Behavior, 29, 515–530. doi: 10.1002/ab.10060.
Merten, D. E. (1997). The meaning of meanness: Popularity, competition,
and conflict among junior high school girls. Sociology of Education,
70, 175. doi: 10.2307/2673207.
Murray, D. M., Rooney, B. L., Hannan, P. J., Peterson, A. V., Ary, D. V.,
Biglan, A., & Futterman, R. (1994). Intraclass correlation among
common measures of adolescent smoking: Estimates, correlates, and
applications in smoking prevention studies. American Journal of
Epidemiology, 140(11), 1038–1050. Retrieved from http://aje.oxfordjournals.
org/content/140/11/1038.full.pdf.
Nansel, T. R., Craig, W., Overpeck, M. D., Saluja, G., & Ruan, W. J.
(2004). Cross-national consistency in the relationship between
bullying behavior and psychosocial adjustment. Archives of
Pediatric Adolescent Medicine, 158, 730. doi: 10.1001/archpedi.
158.8.730.
Noruésis, M. J. (1993). SPSS for Windows SPSS base system for
Windows (Version V6.0.) [Computer program]. Chicago: SPSS, Inc.
O’Connell, P. D. (1999). Peer processes and bullying: Naturalistic
observation on the playground. Unpublished doctoral dissertation,
York University, Toronto, Canada.
O’Connell, P., Pepler, D., & Craig, W. (1999). Peer involvement in
bullying: Insights and challenges for intervention. Journal of
Adolescence, 22, 437–452. doi: 10.1006/jado.1999.0238.
Ohnuma, S., Hirose, Y., Karasawa, K., Yorifuji, K., & Sugiura, J. (2005).
Why do residents accept a demanding rule? Fairness and social benefit
as determinants of approval of a recycling system. Japanese
Psychological Research, 47, 1–11. doi: 10.1111/j.1468–5584.2005.
00267.x.
Olthof, T., Goossens, F. A. (2003, April). Emotional and motivational
correlates of playing a particular role in bullying. Paper session
presented at the Society for Reseach in Child Development Biennial
Meeting, Tampa, Florida.
Pellegrini, A. D. (2002). Bullying, victimization, and sexual harassment
during the transition to middle school. Educational Psychologist, 37,
151–163. doi: 10.1207/s15326985ep3703_2.
Piliavin, J. A. (2001). Sociology of altruism and prosocial behavior. In
N. J. Smelser & P. B. Baltes (Eds.), International encyclopedia of the
social and behavioral sciences, (pp. 411–415) (Oxford, UK: Elsevier
Science.
Pruitt, D. G. (1998). Social conflict, In D. T. Gilbert S. T. Fiske & G.
Lindzey (Eds.), The handbook of social psychology 4th ed., 2, (pp.
470–503). Boston: McGraw-Hill.
Pruitt, D. G., & Kimmel, M. J. (1977). Twenty years of experimental
gaming: Critique, synthesis, and suggestions for the future. Annual
Review of Psychology, 28, 363–392. doi: 10.1146/annurev.
ps.28.020177.002051.
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models:
Applications and data analysis methods, 2nd eds. Thousand Oaks, CA:
Sage.
Salmivalli, C. (1999). Participant role approach to school bullying:
Implications for interventions. Journal of Adolescence, 22, 453–459.
doi: 10.1006/jado.1999.0239.
Salmivalli, C. (2001). Group view on victimization: Empirical findings
and their implications. In J. Juvonen & S. Graham (Eds.), Peer
harassment in school: The plight of the vulnerable and victimized,(pp.
398–419). New York, NY: New York Guilford.
Salmivalli, C., Lappalainen, M., & Lagerspetz, K. M. J. (1998). Stability
and change of behavior in connection with bullying in schools: A twoyear
follow-up. Aggressive Behavior, 24, 205–218. doi: 10.1002/(sici)
1098–2337(1998)24:3 3.0.co;2-j.
Salmivalli, C., & Voeten, M. (2004). Connections between attitudes,
group norms, and behavior in bullying situations. International
Journal of Behavioral Development, 28, 246–259. doi: 10.1080/
01650250344000488.
Siddiqui,O., Hedeker, D., Flay, B. R.,&Hu, F.B. (1996). Intraclass correlation
estimates in a school-based smoking prevention study: Outcome and
mediating variables, by sex and ethnicity. American Journal of
Epidemiology, 144, 425–433. doi: 10.1093/oxfordjournals.aje.a008945.
Snijders, T. A. B., & Bosker, R. J. (1994). Modeled variance in two-level
models. Sociological Methods and Research, 22, 342–363. doi:
10.1177/0049124194022003004.
Stormshak, E. A., Bierman, K. L., Bruschi, C., Dodge, K. A., &Coie, J. D.
(1999). The relation between behavior problems and peer preference
in different classroom contexts. Child Development, 70, 169–182. doi:
10.1111/1467–8624.00013.
Sutton, J., & Smith, P. K. (1999). Bullying as a group process: An
adaptation of the participant role approach. Aggressive Behavior, 25,
97–111. doi: 10.1002/(sici)1098–2337(1999)25:2 3.0.co;2–7.
Tani, F., Greenman, P. S., Schneider, B. H., & Fregosao, M. (2003).
Bullying and the big five: A study of childhood personality and
participant roles in bullying incidents. School Psychology International,
24, 131–146. doi: 10.1177/0143034303024002001.
Tyler, T. R., & Degoey, P. (1995). Collective restraint in social dilemmas:
Procedural justice and social identification effects on support for
authorities. Journal of Personality and Social Psychology, 69, 482–
497. doi: 10.1037/0022–3514.69.3.482.
Van Lange, P.A. M., Liebrand,W.B. G.,Messick,D. M.,&Wilke, H. A. M.
(1992). Introduction and literature review. InW. B.G. LiebrandD.M.
Messick & H. A.M.Wilke (Eds.), Social dilemmas: Theoretical issues
and research findings, pp. vii, 334. Oxford, UK: Pergamon Press.
Xie, H., Swift, D. J., Cairns, B. D., & Cairns, R. B. (2002). Aggressive
behavior in social interaction and developmental adaptation: A
narrative analysis of interpersonal conflicts during early adolescence.
Social Development, 11, 205–221. doi: 10.1111/1467–9507.00195.
Aggr. Behav.
108 Amelia Kohm
Copyright of Aggressive Behavior is the property of John Wiley & Sons, Inc. and its content
may not be copied or emailed to multiple sites or posted to a listserv without the copyright
holder’s express written permission. However, users may print, download, or email articles for
individual use.