Fearful And Distracted In School Article Discussion

Fearful and Distracted in School: Predicting
Bullying among Youths
Steven Lawrence Brewer, Jr., Hannah Meckley-Brewer, and Philip M. Stinson
KEY WORDS: adolescents; children; criminal justice; health and well-being; violence
S
chool-based bullying prevention programs
are consistently implemented in numerous
countries, yet schools remain the most common location for bullying (P. Smith, Pepler, &
Rigby, 2004; Swearer, Espelage, Vaillancourt, &
Hymel, 2010). Bullying is de?ned as any unwanted
aggressive behaviors that occur repeatedly over
time. These aggressive behaviors can be verbal,
physical, or emotional and are accompanied by a
real or imagined imbalance of power ( Juvonen,
2001; Olweus, 1991; P. K. Smith & Brain, 2000).
Victims of bullying often feel disconnected in
school and become less engaged, placing them at
an increased risk for school absenteeism and dropping out (Catalano, Oesterle, Fleming, & Hawkins,
2004; Klem & Connell, 2004; Libbey, 2004). Bullying can even cause children to be fearful of going to
school. Proper youth development is incumbent on
students feeling physically and emotionally safe in
schools.
Children and youths who are bullied are at an
increased risk for low self-esteem and mistrust of
others (Eckholm & Zezima, 2010). Given evidence
that bullying is associated with problems in child and
adolescent development for both bullies and victims,
mitigating bullying is critical for youths to transition
e?ectively into adulthood (Shaheen, Nassar, Saleh, &
Arabia, 2014; Warren, 2011). Bullying is linked to
anxiety, depression, school refusal, and?most
doi: 10.1093/cs/cdx021
? 2017 National Association of Social Workers
concerning?youth suicide (Bethune, 2014; Turner,
Exum, Brame, & Holt, 2013; Williams & Kennedy,
2012). This topic is of timely importance as youth
suicide is the second leading cause of death among
15- to 29-year-olds worldwide (World Health Organization, 2016).
As many factors a?ect the likelihood of bullying
in schools, there is no ?one size ?ts all? strategy to
bullying prevention, and bullying prevention requires a ?whole school? approach (Birch & Videto,
2015; P. Smith et al., 2004). Extant research suggests
that students? fear of being attacked, avoidance,
lower grades, and the level of school security are all
predictors for bullying (Gutt & Randa, 2016;
Wynne & Joo, 2011). Bullying has been a topic of
interest for several decades, yet continues to a?ect
approximately 20 percent of school-age youths
(Warren, 2011). Although crime rates among the
general population have gone down, bullying
among school-age youths has not diminished
(Robers, Kemp, Rathbun, A., & Morgan, 2014).
As such, it is imperative to further examine whether
additional behaviors or characteristics also predict
bullying in schools (Tanner-Smith & Fisher, 2016).
In this study, we used chi-square automatic interaction detection (CHAID) decision tree and logistic
regression models to identify risk factors that
increase bullying in schools in the United States.
We hypothesized that fear, avoidance, lower grades,
219
Downloaded from https://academic.oup.com/cs/article/39/4/219/4098141 by guest on 01 February 2022
Bullying and aggression in schools can have a traumatic and lasting e?ect on the well-being
of children and youths. Using data from the 2013 National Crime Victimization Survey?s
School Crime Supplement, this study uses a chi-square automatic interaction detection
(CHAID) decision tree and logistic regression models to identify factors that increase bullying in schools. Being distracted and fear of being attacked were among the top statistically
signi?cant variables, when using both methodologies. Avoiding online activities and knowing someone who brought a gun to school were top predictors using logistic regression.
Being involved in a ?ght and seeing hate-related words or symbols were additional in?uential predictors in a CHAID decision tree model. Identifying factors that increase likelihood
for bullying in schools assists practitioners in implementing programs and policies to improve school climate and reduce youth bullying.
the presence of weapons, and school security measures would be identi?ed as signi?cant predictors of
bullying, but that using two complementary statistical methods may identify additional signi?cant variables for bullying prediction.
METHOD
220
ANALYTIC PROCEDURES
At the bivariate level, chi square measures the statistical signi?cance of the association between two variables. Cram?r?s V measures the strength of that
relationship with values that range from 0 to 1.0 and
allows for an ?assessment of the actual importance of
the relationship? (Ri?e, Lacy, & Fico, 2005, p. 191).
To improve predictive power regarding bullying,
two distinct but complementary analyses were conducted. First, a forward stepwise binary logistic
regression model was used. Stepwise binary logistic
regression is used to determine which of the independent variables are statistically signi?cant in a multivariate model with a dichotomous dependent
variable. Stepwise logistic regression models are
appropriate when the study is purely exploratory
and predictive (Menard, 2002). Regression diagnostics are reported for evaluation of the logistic
regression model.
Second, classi?cation tree analysis?also known as
a decision tree?was used to uncover the decision
rules that describe the relationship between independent predictors and bullying. This approach moves
beyond the simple one-way additive relationship of
linear statistical models by identifying the hierarchical interactions between the independent predictors
and their compounding impact. Classi?cation trees
examine the entire data set and produce a graphical
output that ranks the variables by statistical importance. The most in?uential variable (the root node)
is represented at the top of the tree. This variable is
used to split the data in a recursive manner through
the creation of subsets into the lower branches of the
tree. Variable selection and splitting criteria are then
driven by the algorithm of the tree program. Decision
tree techniques have received attention because of
their ability to handle interaction e?ects in data without being bound to statistical assumptions (Sonquist,
1970). Classi?cation tree analysis has been used to
examine cost-e?ectiveness of the Olweus Bullying
Prevention Program (Beckman & Svensson, 2015),
ecological factors of being bullied among adolescents (Moon, Kim, Seay, Small, & Kim, 2015), and
a machine learning approach to detect cyberbullies
(Dadvar, Trieschnigg, & de Jong, 2014).
We used CHAID, a classi?cation technique
that uses the p value of the chi square for splitting
criteria (Kass, 1980). CHAID examines all possible
Children & Schools Volume 39, Number 4 October 2017
Downloaded from https://academic.oup.com/cs/article/39/4/219/4098141 by guest on 01 February 2022
Data were obtained from the 2013 National Crime
Victimization Survey (NCVS) School Crime Supplement (SCS) (National Archive of Criminal Justice Data, 2014). The NCVS is an annual survey
administered by the U.S. Census Bureau for the
Bureau of Justice Statistics to obtain information
about school-related safety. The SCS is a national
interview of approximately 6,500 students ages 12
through 18 attending public and private schools in
the United States. Speci?cally, the SCS investigates
students? perceptions of crime and safety at school
and their experiences with bullying.
Data for the 2013 SCS were collected through
phone or face-to-face interviews. For the current
study, casewise deletion was used to create a sample with completed data based on the measure of
bullying (N = 4,967). The sample for the current
study was 51.5 percent male and 77.5 percent
white, with a mean age of 14.95 years. Ninetytwo percent attended public school, and 40.5 percent had household incomes greater than $50,000.
Based on prior research, the independent variables selected for the current study were transportation to school, participation in extracurricular
activities, school security measures, student perception of school rules, student perception of teachers
and adults in school, crime in neighborhood
around school grounds, drug availability, avoidance behaviors, fear of attack or harm, presence or
knowledge of weapons at school, and presence of
gangs at school. The control variables included
gender, race, school type, and household income.
The dependent variable for the current study was
bullying. Bullying was measured by the respondent?s answers to seven questions. A scale bullying
variable was created from the seven questions, and
Cronbach?s alpha coe?cient yielded a reliability
score of .736. The questions asked: Did someone
(1) make fun of you, called you names, or insulted
you, in a hurtful way? (2) spread rumors about you
or tried to make others dislike you? (3) threaten you
with harm? (4) push you, shove you, trip you, or
spit on you? (5) try to make you do things you did
not want to do? (example: give them money or
other things), (6) exclude you from activities on
purpose? (7) destroy your property on purpose?
RESULTS
The analysis identi?ed that 21.7 percent of the respondents (n = 1,097) were bullied during the six
months prior to the 2013 survey. Of those, 67 percent of the students were bullied only one or two
times per year and 19.3 percent of students were bullied one or two times per month. Of the population
interviewed, 7.6 percent were bullied one or two
times per week, and 6 percent reported being a victim of bullying almost every day. Ninety-two percent of students included in the study were attending
public school. Cases were geographically distributed
across the United States as follows: Northeast 15.8
percent, Midwest 24.2 percent, South 36 percent,
and West 23.9 percent. Approximately 80 percent
were from metropolitan settings and 19.8 percent represented nonmetropolitan settings.
Bivariate chi-square associations were statistically signi?cant at p < .05 for 71 independent variables and the dependent variable, bullying (where
0 = no and 1 = yes). Eight of the bivariate associations were of moderate strength as indicated by
the Cram?r?s V score. They include how often a
student is distracted, where ?2(3, N = 4,948) =
453.858, p < .001, V = .303; presence of gangs
that sell drugs, where ?2(1, N = 627) = 39.939,
p < .001, V = .252; afraid of attack or harm at
school, where ?2(3, N = 4,958) = 306.790, p < .001,
V = .249; how often were gangs involved in violence, where ?2(5, N = 1,423) = 82.084, p < .001,
V = .240; knew students were on drugs or alcohol
during school, where ?2(1, N = 4,925) = 269.976,
p < .001, V = .234; someone destroyed their property during school, where ?2(1, N = 4,965) =
270.542, p < .001, V = .233; saw hate words written in school, where ?2(1, N = 4,944) = 231.918,
p < .001, V = .217; o?ered drugs beside alcohol
and tobacco during school, where ?2(1, N = 4,958) =
228.331, p < .001, V = .215.
Table 1 presents a forward stepwise binary logistic regression model predicting bullying. Computation of bivariate correlations for each of the
independent variables in the model revealed that
none of the variables were highly correlated with
each other. Multicollinearity was not a problem as
indicated by no tolerance scores below .954 and no
variance in?ation factors above 1.048. The Durbin?
Watson score of 1.885 indicated that autocorrelation was not a problem in the model. Logistic
regression results indicated that the overall model
of eight predictors was statistically reliable in distinguishing between students who were bullied and
those who were not bullied. The binary logistic
regression model correctly classi?ed 80.3 percent of
the cases (AUC = .722, 95% con?dence interval
[CI] [.705, .740], R2ROC = .444). Wald statistics
indicated that all of the independent variables in
the model signi?cantly predicted bullying.
Interpretation of the odds ratios in the binary
logistic regression model provides context for prediction of bullying. Seven of the independent variables
in the regression model predict when bullying is
more likely to occur. The simple odds that a student will be bullied are two times greater if that
student is easily distracted in school. Distraction
is a signi?cant predictor of bullying among the
students surveyed. Next, the simple odds that a
student will be bullied are 13 times greater if that
student avoids online activities. Avoidance of
speci?c activities or locations is often indicative
of bullying. The simple odds that a student will
be bullied are 14 times greater if that student
knows someone who has brought a gun to
school, and the simple odds that a student will be
bullied are 18.2 times greater if that student has
brought a knife to school.
Brewer, Meckley-Brewer, and Stinson / Fearful and Distracted in School: Predicting Bullying among Youths
221
Downloaded from https://academic.oup.com/cs/article/39/4/219/4098141 by guest on 01 February 2022

splits for predictor variables and selects predictors
based on the optimal number of splits that can be
created. CHAID was performed with the following parameters: (a) tree depth set to ?ve levels, (b)
parent nodes limited to no less than 10 cases, (c)
child nodes limited to no less than ?ve cases, and
(d) likelihood ratio chi-square splitting criteria.
The predictive power of logistic regression and
classi?cation trees was assessed through the area
under the curve (AUC) component of the receiver
operating characteristic (ROC). The AUC assesses
the predictive accuracy of a statistical model and
serves as the preferred method for assessing and
comparing models (Bewick, Cheek, & Ball, 2004;
Dolan & Doyle, 2000). The ROC curve considers
the sensitivity versus 1? speci?city, a representation
of the true positive rate (TPR) versus the false positive rate (FPR). The curve is displayed graphically
by plotting the TPR on the y axis and the FPR on
the x axis. ROC curves are interpreted through
the AUC, a score that ranges from 0 to 1. A straight
line through a ROC curve is the equivalent of .5
and suggests that the model is no better at prediction than ?ipping a coin. A score of 1 indicates that
the model is able to accurately predict all cases.
Table 1: Logistic Regression Model Predicting Bullying (N = 143)
Variable
B
SE
Wald
p
Exp(B)
95% CI for Exp(B)
1.121
2.645
2.721
1.016
1.161
0.859
2.957
?1.044
196.382
96.965
.353
.482
.444
0.260
1.067
0.703
0.369
0.559
0.236
1.463
0.338
18.634
6.147
14.995
7.566
4.322
13.235
4.084
9.551
.000
.013
.000
.006
.038
.000
.043
.002
3.067
14.085
15.191
2.761
3.194
2.361
19.237
0.352
[1.844, 5.101]
[1.740, 113.996]
[3.833, 60.205]
[1.339, 5.694]
[1.069, 9.543]
[1.486, 3.751]
[1.093, 338.549]
[0.182, 0.683]

Purchase answer to see full
attachment