Child Abuse & Neglect 39 (2015) 109–122
Contents lists available at ScienceDirect
Child Abuse & Neglect
The association between school exclusion, delinquency and subtypes of cyber- and F2F-victimizations: Identifying and predicting risk profiles and subtypes using latent class analysis Gia Elise Barboza ∗ African American Studies and Criminology, Northeastern University, 200F Renaissance Park, 360 Huntington Ave., Boston, MA 02115, USA
a r t i c l e
i n f o
Article history: Received 1 March 2014 Received in revised form 6 August 2014 Accepted 8 August 2014 Available online 5 September 2014 Keywords: Cybervictimization Bullying Academic performance Aggression Latent class analysis
a b s t r a c t This purpose of this paper is to identify risk profiles of youth who are victimized by onand offline harassment and to explore the consequences of victimization on school outcomes. Latent class analysis is used to explore the overlap and co-occurrence of different clusters of victims and to examine the relationship between class membership and school exclusion and delinquency. Participants were a random sample of youth between the ages of 12 and 18 selected for inclusion to participate in the 2011 National Crime Victimization Survey: School Supplement. The latent class analysis resulted in four categories of victims: approximately 3.1% of students were highly victimized by both bullying and cyberbullying behaviors; 11.6% of youth were classified as being victims of relational bullying, verbal bullying and cyberbullying; a third class of students were victims of relational bullying, verbal bullying and physical bullying but were not cyberbullied (8%); the fourth and final class, characteristic of the majority of students (77.3%), was comprised of non-victims. The inclusion of covariates to the latent class model indicated that gender, grade and race were significant predictors of at least one of the four victim classes. School delinquency measures were included as distal outcomes to test for both overall and pairwise associations between classes. With one exception, the results were indicative of a significant relationship between school delinquency and the victim subtypes. Implications for these findings are discussed. © 2014 Elsevier Ltd. All rights reserved.
Introduction The upsurge in school shootings during the past decade has been paralleled by a dramatic increase in research on the criminogenic effects of peer victimization at school (Juvonen & Graham, 2001; Garbarino & DeLara, 2004; Rubinlicht, 2011; Hong & Espelage, 2012). The impact that bullying has on future criminal behavior and/or delinquency has made it a particularly fruitful area of investigation. While bullying typically conjures up images of physical intimidation, the reality is that youth are victimized in a variety of ways, any or all of which may result in irreparable psychological and/or physical harm (Cullen, Unnever, Hartman, Turner, & Agnew, 2008). In addition to overt acts of hostility, covert forms of harassment
∗ Corresponding author. http://dx.doi.org/10.1016/j.chiabu.2014.08.007 0145-2134/© 2014 Elsevier Ltd. All rights reserved.
110
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
including spreading rumors or lies and excluding and/or ignoring other students from school activities are commonplace. Moreover, as a result of today’s electronic age, the Internet is supplanting traditional forms of schoolyard bullying however victims characteristics differ according to the type of harm being inflicted (Law, Shapka, Hymel, Olson, & Waterhouse, 2012). The differences in form and function between traditional bullying and cyberbullying begs for consideration of whether this “new” form of harassment (Wade & Beran, 2011) that uses “email, instant messages, cell phones, text messages, photos, videos and social networking websites to humiliate and threaten others” (Grim, 2008, p. 157) is really just a new way to implement the same behavior or rather whether it is indeed a distinct phenomenon. Despite significant advances in our understanding of bully victimization over the past decade, much remains to be understood. An important area of research is to examine potential overlaps in harassment and bullying across all possible environments to gain understanding of the sum total of youth’s experiences with victimization (Ybarra, Diener-West, & Leaf, 2007). For example, what victim profiles emerge when multiple indicators of both online and offline harassment are analyzed? Also, what is the association between victim profiles and delinquent behavior at school such as skipping class, absenteeism and/or weapon-carrying and physical fighting? The aim of this paper is to document the co-occurrence of online and face-to-face peer harassment and bullying and its effect on several measures of school alienation, avoidance and delinquency. Latent class analysis is used to identify the nature and form of victimization and bullying in early adolescence, explore the overlap and co-occurrence among different clusters of victims, and to examine the relationship between multiple risk factors for school exclusion, delinquency and membership in each ‘victim’ class. Prevalence of bullying and cyberbullying Aggression includes any single act of physical, emotional and/or psychological harm including physical or sexual violence, psychological or emotional abuse or neglect (U.S. Department of Justice, 2004). Common to most definitions of bullying, on the other hand, is that it is repetitive, results in harm and occurs in the context of an imbalance of either psychological or physical power. While fights, one-time attacks, or “harmless” teasing are appropriately considered aggressive behaviors (Berger, 2007), individuals subjected to these same behaviors repeatedly are considered bullied in a more general sense. Examples of bullying include being called names, being physically hurt, being threatened, being the subject of rumors, being socially isolated and having one’s belongings taken repeatedly (Glew, Rivara, & Feudtner, 2000). Similar to traditional bullying, cyber bullying is often defined as “willful and repeated harm” (Hinduja & Patchin, 2010, p. 5) inflicted toward another. In contradistinction to traditional face-to-face bullying, however, cyberbullying utilizes electronic communication to threaten, harass embarrass, or socially exclude (Hinduja & Patchin, 2010; Patchin & Hinduja, 2006; Williams & Guerra, 2007). While there is no universally accepted definition of cyberbullying (Smith, del Barrio, & Tokunaga, 2013), the very nature of the Internet as being anonymous, accessible and wide-reaching implies that all the elements of bullying are present in a single online interaction. Cyberbullying involves repetition not only because material such as email, text, or pictures can be viewed by anyone with online access (Campbell, 2005; Slonje & Smith, 2008) but also because the same communication can be repeated a number of times through mass communication. The anonymous nature of the interaction endows the perpetrator with power regardless of his or her personal characteristics. The difficulty associated with removing unwanted or harmful communication as well as its permanency only serves to reinforce the power and repetitiveness of the interaction (Wolak, Mitchell, & Finkelhor, 2007). A large body of research documenting the prevalence, magnitude and impact of peer-to-peer victimization challenges commonly held assumptions that school is a place where children are safe. Nansel, Overpeck, and Pilla (2001) at the National Institute of Child Health reported that bullying affects approximately 30% of adolescents in the United States, 13.6% as bullies, 10.6% as victims, and 6.3% as bully-victims. While estimates differ greatly due to different methodologies and conceptualizations of “bullying,” most scholars agree that between approximately 15–20% of students are regular victims of bullying behavior while about 10% of children in the US experience “extreme victimization” by bullying (Seals & Young, 2003; Peskin, Tortolero, Markham, Addy, & Baumler, 2007). The most prevalent form of bully victimization is name-calling (13.5%) followed by physical aggression (10.8%), teasing (9.0%), exclusion (7.3%) and threats of harm (6.6%) (Seals & Young, 2003). Characteristics of bully and cyberbully victims Initial research on bullying reported that physical bullying declines as youth get older (Perry, Kusel, & Perry, 1988), however later research clarified the existence of a parabolic relationship between physical bullying and age. Bullying is less common among younger and older students, peaking sometime during elementary school (Borg, 1998) around age 7, and declines steadily thereafter through the age of 15 (Glew et al., 2000). Although physical bullying tends to lessen over time, verbal abuse continues to remain high and to be associated with physical abuse as children age (Perry et al., 1988) Type of bully victimization differs according to gender. Girls are more often the targets and perpetrators of passive, indirect bullying, also known as “relational” bullying, such as being the object of gossip and social isolation. Boys on the other hand are more commonly the targets and perpetrators of aggressive, physical bullying (Glew et al., 2000). More recent studies confirm that boys experience physical bullying to a greater extent than girls, the latter being more likely to report being teased or joked about (Carbone-Lopez, Esbensen, & Brick, 2010). Similarly, Dukes, Stein, and Zane (2010) found that adolescent girls reported more relational victimization and adolescent boys reported more physical bullying and victimization, more weapon carrying, and more injury.
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
111
Consistent with the literature on victimization, some researchers have found that the prevalence of victimization by bullying is higher among racial and ethnic minority youth, but results are inconsistent (Peskin et al., 2007) and little is known about racial/ethnic differences in bullying risk (Spriggs, Iannotti, Nansel, & Haynie, 2007). Seals and Young (2003) found no significant differences in the prevalence of victimization between African American and white students. More recently, researchers have documented racial/ethnic differences in bully prevalence, with black students reporting less victimization than both whites and Hispanics (Spriggs et al., 2007). On the other hand, Nansel et al. (2001), who used a representative national sample of youth in grades 6 through 10, found that black youth reported being bullied significantly more than either whites or Hispanics. In yet a different study, Hispanic students reported less victimization than white or African American students however, when they were victimized, they were more likely to experience repeat victimization (Hanish and Guerra, 2002). Clearly, more work needs to be done in order to partial out the effects of race and ethnicity on bully victimization. Schools as correlates and consequences of bullying and cyberbullying Existing research has consistently shown that the interpersonal and institutional settings in which student interactions take place play an important role in the psychosocial functioning of students. Researchers conceptualize bullying (Craig & Pepler, 2007) and cyberbullying (Hoff and Mitchell, 2009; Mishna, Saini, & Solomon, 2009) as part of the broader context of embedded socio-ecological relationships. Accordingly, theorizing about bullying behaviors in children and adolescents implicates both individual and contextual factors (Barboza et al., 2009). The school environment, in particular, is a place where students’ beliefs about violence are formed and where students learn to model the behavior of their peers and adults (Osterman, 2000). Therefore, schools are key in helping us understand the complexity of bullying and for developing sensitive and effective interventions (Barboza et al., 2009; Limber, 2006; DeLara, 2006; Garbarino & DeLara, 2004; Bronfenbrenner and Morris, 1998). Bullying, cyberbullying, internalizing behavior at school Children who are victimized are more likely to report increased levels of internalizing symptoms, such as depression, suicidal ideation and loneliness, not only in middle (Seals & Young, 2003) and high school (Peskin et al., 2007) but for years afterwards. Early research focused on the effect of school size, class size and competition at school on subsequent bullying but attention later shifted to explore different aspects of the school environment such as alienation and school-related stress. Previous research has shown that a wide variety of school characteristics are associated with the presence of both bullying and cyberbullying. These include school-related stressors (e.g. unreasonable expectations of student performance), teacher apathy and/or lack of support, school atmosphere (e.g. feelings of belonging, fair rules) and the broader peer environment including both the presence of peers (Atlas & Pepler, 1998; Craig & Pepler, 2007) and the nature of peer-to-peer relationships (Barboza et al., 2009). One investigation found that positive disciplinary actions and high levels of adult supervision, school involvement and academic achievement characterize schools reporting low levels of bullying. Using grade point average as an objective measure of academic achievement, Glew, Fan, Katon, and Rivara (2008) reported that victims have significantly lower academic achievement in school than non-victims. In sum, research supports the claim that victims have lower levels of affective attachment to school: they are neither as involved nor as committed to school compared to non-victims (Spriggs et al., 2007). In comparison to non-victimized youth, cyberbullied youth report more sadness, anxiety and fear (Beran & Li, 2008); have difficulty with social interactions (Ybarra, Mitchell, Wolak, & Finkelhor, 2006; Blais, 2008); are more likely to be engaged in risky behavior such as drug and alcohol use (Mishna, Cook, Gadalla, Daciuk, & Solomon, 2010) and eating disorders (DeHue, Bolman, & Völlink, 2008; Fosse & Holen, 2006). On this basis, social and emotional functioning is likely to have a mediating impact on the relationship between bullying and educational outcomes. According to one study, 7% of America’s 8th graders stay home at least one day a month because they are afraid of other children at school. Fear of school, in turn, is associated with lower self-esteem and poor school performance. Research has shown that after adjusting for age, sex, ethnicity and income, victims of traditional bullying are significantly more likely than bystanders and non-victims to feel unsafe at school and more likely to feel like they “do not belong” at school (Glew et al., 2008). Students who are bullied in cyberspace only, and students bullied both in cyberspace and at school, experience difficulties at school such as low marks, poor concentration and absenteeism (Beran & Li, 2008). Moreover, they are less able to concentrate at school (Beran & Li, 2005; Juvonen & Gross, 2008), which most likely results in the greater likelihood of skipping school (Ybarra et al., 2007), and receiving detentions and suspensions (Wolak et al., 2007; Ybarra et al., 2007) among victims. Differences in frequency and type of victimization may be important determining factors for predicting compromised psychosocial ability and academic performance characteristic of abused children. Bullying, cyberbullying and externalizing behavior at school. There is a clear relationship between being the victim of bullying and subsequent aggression. As scholars have begun to explore the link between bully victimization and aggressive behavior (Anderson et al., 2001), many have found that a subset of bullied youth respond aggressively to their victimization while others respond more “passively.” Passive victims have greater feelings of insecurity, a negative worldview and are more like to view themselves as a failure in comparison to non-victims (Glew et al., 2000). Provocative victims, on the other hand,
112
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
are not afraid of confrontation. As a result, provocative victims are at greater risk of engaging in violent behavior. Research has shown that bully-victim relationships underlie the majority of childhood assaults, suicides and homicides with bullyvictims at risk for committing suicide and/or murder. Anderson et al. (2001) studied school-associated violent deaths in the US between 1994 and 1999 and found that homicide perpetrators are more than twice as likely as homicide victims to have been victims of bullying at school. Several additional studies since this seminal piece have confirmed the relationship between weapon-carrying and bully victimization. For example, youth who reported being targeted by Internet harassment were eight times more likely than all other youth to concurrently report carrying a weapon to school (Ybarra et al., 2007). An analysis of 15 school shootings the occurred between 1995 and 2001 revealed that in 87% of the cases, shooters had experienced relational bullying, especially chronic rejection (Leary, Kowalski, Smith, & Phillips, 2003, Dukes et al., 2010). Using data from 1,300 adolescents in a Colorado school district, Dukes et al. (2010) found that being a victim of physical bullying, a victim of relational bullying, or being a relational bully significantly predicted more injury among both boys and girls. The devastation that results from the confluence of victimization by bullying, weapon-carrying and social withdrawal is permanently established in the public’s consciousness due to recent school massacres occurring at elementary, high schools and universities across the country. Bully and cyberbully typologies and associated risk profiles The terms “bullying” and “cyberbullying” encompass a variety of different behaviors and experiences (Cornell & Mehta, 2011; Lovegrove & Cornell, 2013) that may or may not be independent of each other (Wang, Iannotti, Luk, & Nansel, 2010). Bully victimization is multifaceted and includes verbal, physical and/or psychological aggression. An emergent body of research documenting the overlap and co-occurrence of these behaviors exemplifies their heterogeneity. For example, although physical and verbal harassment are conceptually distinct, empirically victims are more likely to be verbally harassed given that they have experienced physical abuse. Prior research has shown that as many as 58% of students report being victims of both verbal and physical bullying (Orpinas, Horne, & Staniszewski, 2003). In addition, the experiences of frequently victimized children fall along distinct dimensions (Wang et al., 2010) each characterized by different coping strategies and adjustment problems (Waasdorp & Bradshaw, 2011). For example, researchers have found distinct classes of victims including an ‘all-types’ victim class (9.7% of males and 6.2% of females), a verbal/relational victims class (28.1% of males and 35.1% of females), and a non-victim class (62.2% of males and 58.7% of females) (Wang et al., 2010). A recent study conducted by Lovegrove and Cornell (2013) used latent class analysis to study the association between bullying involvement and internalizing and externalizing problem behaviors among over 3,500 high school students from Virginia. They found four latent classes of bullies and victims composed of a non-involved class (65%), a bullies class (12%), a victims class (16%), and a bully-victims class (8%). Externalizing problem behaviors were significantly higher among students in the bullies and bully-victims classes, while internalizing problem behaviors were higher among victims and bully-victims. These analyses directly or indirectly point to the fact that both polyvictimization and the frequency of abuse has more adverse effects on internalizing and externalizing symptoms such as psychological well-being, academic achievement and criminogenic effects (Berger, 2007; Buhs, Ladd, & Herald, 2006; Hawker & Boulton, 2000; Nansel et al., 2001). The current study The present study expands upon previous work in several important ways. First, unlike previous studies that model bully victimization as a random and independent event, the present analysis models the interdependent nature of victimization by allowing for the possibility that more than one form of victimization is part of the bully-victim experience. Secondly, it is one of the first studies to utilize latent class analysis to explore the co-occurrence of cyberbullying and bullying victimizations in a nationally representative sample using five separate indicators of both types of behavior (10 all together) (but see Wang et al., 2010). In the analysis that follows, conditional class means are used to describe the latent structure of bullying – both traditional and online – among youth. The current study also addresses the question of which victimhood classe(s) impacts school participation and delinquency, including weapon carrying and fighting, the most—a question not previously addressed in the existing literature. To do so, only academic outcomes that have directly resulted from students’ own experiences with victimization are utilized. Analytically, the latent classes are used as predictors of school outcomes and are tested for significant associations. This allows for the possibility of quantifying the impact of victimization across several academic indicators, providing vital information regarding potential targets for the prevention of academic and behavior problems in children. More specifically, this paper addresses the following research questions: 1. What is the prevalence and co-occurrence of online and face-to-face victimization by bullying; 2. How can we describe the risk profiles of victims? Do victims of bullying and cyberbullying have distinct profiles (i.e. is there a separate class of cyberbullied and bullied) or are they both part of the same phenomenon (i.e. are victims of bullying likely to be cyberbullied and vice versa). In order to address this question, a latent class analysis of victims was implemented to predict class prevalence as well as the conditional probability of being a victim latent class membership; 3. How do profiles of risk vary with victims’ sociodemographic characteristics? A multinomial logistic latent class regression was performed using characteristics such as age, gender, race, and grade to predict latent class membership.
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
113
4. Does membership in a victim class predict school behavior? Posterior probabilities were used to predict latent class membership and then pairwise comparisons were performed for the purpose of testing the equality of means across 6 measures of school behavior (including skipping classes, being absent from school, fighting at school and carrying a weapon to school). Methods Participants Data for this study were derived from the 2011 National Crime Victimization Survey: School Crime Supplement (SCS). Twice each year, data are obtained from a nationally representative sample of roughly 49,000 households comprising about 100,000 persons on the frequency, characteristics, and consequences of criminal victimization in the United States. Youth between 12 and 18 years old living in a household selected to complete the NCVS are eligible for inclusion in the SCS. The final sample includes persons between 12 and 18 years of age who were in primary or secondary education programs leading to a high school diploma (elementary through high school), and who had been enrolled sometime during the 6 months prior to the interview. The SCS asks questions related to students’ experiences with, and perceptions of, crime and safety at school, student bullying, and fear of victimization at school. Mplus 6.12 (Muthen & Muthen, 1998–2010) was the software used to analyze data for the current study.
Measures Demographic variables Demographic variables included age, gender, grade (5th–12th) and race which was coded as white/non-white irrespective of ethnicity.
Peer victimization Seven items from the SCS were utilized to measure students’ experiences with bully victimization. The participants were asked the following, “Now I have some questions about what students do at school that make you feel bad or are hurtful to you. We often refer to this as being bullied. You may include events you told me about already. During this school year, has any student bullied you?” The responses were coded as either direct or indirect bullying. In accordance with previous literature, direct bullying refers to situations in which the victim is directly involved whereas indirect bullying may take place without the victims immediate knowledge or presence (Vandebosch & Van Cleemput, 2009). Therefore, direct bullying included physical bullying (“Pushed you, shoved you, tripped you, or spit on you?”), verbal bullying (“made fun of you, called you names, or insulted you, in a hurtful way”), being threatened with harm, being coerced into doing things you don’t want to do, being excluded from activities on purpose and having property destroyed. Indirect bullying was measured with one item, namely tapping whether the respondent was the object of rumors or other forms of communication intended to make others dislike them (“spreading rumors about you or tried to make others dislike you”). Students’ experiences with cyberbullying were measured with four items that were similarly classified as either direct or indirect. Direct cyberbullying included threats of harm made online (“whether another student used the internet to threaten or insult you”), online exclusion (“being purposefully excluded from online communications”), and non-verbal harm (“someone posted hurtful information about you on the Internet”), while indirect bullying was measured as outing (“purposefully sharing your private information, photos or video”).
School alienation Four items were used to measure school withdrawal/avoidance. Participants were asked the following questions: (1) “during this school year, did you STAY AWAY from any online activities because you thought someone might be mean to you there?” (2) “Did you AVOID any activities at your school because you thought someone might attack or harm you?” (3) “Did you AVOID any classes because you thought someone might attack or harm you?” and (4) “Did you stay home from school because you thought someone might attack or harm you in the school building, on school property, on a school bus, or going to or from school?”
Aggressive behaviors at school. Two items measured aggressive behavior at school. Students were asked, “During this school year, have you been in one or more physical fights at school?” and During this school year, did YOU ever bring the following to school or onto school grounds?” “A gun?” All response choices were binary (1 = Yes/2 = No).
114
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
Fig. 1. A latent class mixture model of traditional and cyberbullying with academic risk as a distal outcome and youth demographic characteristics as predictors of class membership.
Statistical analyses Statistical analyses were conducted using MPLUS 6.2. First, descriptive statistics of all variables were examined. Second, a series of latent class models were estimated (1) to explain the relationships among a set of observed measures of victimization by means of discrete but latent groups; (2) to identify “types” or classes of victims; and (3) to derive an estimate of the prevalence of each class. The first step in LCA Competing solutions were compared on the basis of model fit indices, interpretability, and parsimony. Determination of model fit was based on log-likelihood, Akaike Information Criteria (AIC), Bayes Information Criteria (BIC), sample size adjusted Bayes Information Criteria (a-BIC), entropy, and Lo–Mendel–Rubin adjusted likelihood ratio test (LMR) (Lo et al., 2001). Smaller values of log-likelihood, AIC, BIC and a-BIC indicate better fit to the data whereas significant p-values associated with an LMR indicate significant improvement in model fit relative to the solution with one less class. Once a final solution was identified, the model was adjusted for age, gender, race and grade level. After the final model was estimated, auxiliary variables measuring school alienation/withdrawal and aggressive behavior were used as outcome variables. This approach removed the auxiliary variables’ influence over the latent structure and hence did not change the substantive interpretation of the classes (Wang et al., 2010). The analytical model is represented in Fig. 1.
Results Sample characteristics The sample consists of 5,589 youth between the ages of 12 and 18 (mean = 14.77, s.d. = 1.99). Their highest level of academic achievement was the 9th grade, on average. Males and females comprised 51% and 49% of the sample, respectively. The majority of respondents identified racially as “white” (80%) and the rest as non-white (20%) including African/American or black, Hispanic and/or Asian.
Descriptive characteristics Prevalence of cyber and bully victimization. Descriptive statistics of the bullying behaviors are provided for the whole sample and by gender in Table 1. Table 1 shows the distinction between direct and indirect forms of bullying. The prevalence of victimization from the 7 types of face-to-face bullying were 18.5% for spreading rumors, 17.7% for verbal (males: 17.1, females: 19.7), 7.9 for physical (males: 9.5, females 7.1), 5.5% for social exclusion (males: 4.5, females 6.5), 5.1 for threatened with harm (males: 5.5, females 5.1), 3.3% for being forced to do things (males: 3.5, females: 3.0) and 2.8% for destroyed property (males: 3.5, females: 2.2). Being threatened or insulted online, via text messaging or via gaming (i.e. verbal cyberbullying) was the most common form of cyberbullying (7.0%; n = 237) followed by non-verbal cyberbullying (subjected to the posting of hurtful information; 3.7%; n = 201), social exclusion (purposefully being excluded from online communications; 1.2%; n = 64) and then outing (having private information, photos or videos on the internet or via phone shared in a hurtful way (1.1%; n = 58). The single measure reported here of being threatened is a constellation of three variables: online threats, threats received from text messaging and on threats received via gaming websites. In the analyses that follow, being threatened by text message was kept as a separate variable but gaming and online threats were consolidated. Consequently, there are 11 victimization variables in Table 1 but 12 victimization variables reported in Table 3.
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
115
Table 1 Prevalence of bullying and cyberbullying behaviors and chi-square tests of association by gender 1. Prevalence of traditional and cyber victimizations among peers by gender. Traditional Victimization 1. Direct Verbal (i.e. calling someone names) Threatened w/harm Physical (i.e. hitting, pushing) Forced to do things Social (i.e. excluding someone from group activities) Property (i.e. destroying someone’s property) 2. Indirect Spread false rumors Cybervictimization 1. Direct Verbal (i.e. using internet to threaten or insult someone) Social (i.e. excluding someone from online communications) Non-verbal (i.e. posting hurtful information) 2. Indirect Outing (i.e. sharing personal information)
Males
Females
Total
Odds ratio
17.1 5.2 9.5 3.5 4.5 3.5
19.7 5.3 7.1 3.0 6.5 2.2
17.7 5.1 7.9 3.3 5.5 2.8
1.19*** 1.02 0.728*** 0.853 1.48*** 0.620***
12.8
24.5
18.5
2.21***
5.4 0.9 1.8
8.8 1.5 5.9
7.0 1.2 3.7
1.69*** 1.68** 3.42***
0.7
1.5
1.1
2.16***
Table 2 Outcome measures: school withdrawal and avoidance due to fear of being attacked, and aggressive behaviors at school. Social withdrawal from/avoidance of school
n (%)
Stay away from online activities Avoid activities at school Skip classes at school Stay home from school Aggressive behaviors at school Been in one or more physical fights Carried a gun to school
108 (2.0) 67 (1.2) 43 (0.7) 50 (1.0) 262 (4.5) 42 (0.7)
Table 3 Model Fit Statistics for the n-class models of peer victimization. Akaike (AIC)
Bayesian (BIC)
Sample size adjusted BIC (a-BIC)
Entropy
Lo, Mendell, Rubin test
n for each class
1-LC 2-LC 3-LC 4-LC
28,693.177 22,786.01 22,396.149 22,209.327
28,773.16 22,952.641 22,649.429 22,549.256
28,735.027 22,873.199 22,528.676 22,387.193
na 0.899 0.883 0.885
na 5,871.567 412.201 210.792
5-LC
22,127.912
22,554.489
22,351.116
0.859
106.393
C1 = 5,798 C1 = 4,770; C2 = 1,028 C1 = 4,773; C2 = 816; C3 = 209 C1 = 4,763; C2 = 662; C3 = 242; C4 = 131 C1 = 4,507; C2 = 741; C3 = 284; C4 = 163; C5 = 103
The results show that bully victimizations differ significantly by gender. The odds of being bullied by rumors or being subjected to the unwanted posting of hurtful information, for example, are 2.11 (p < 0.000) and 3.58 (p < 0.000) times higher for females compared to males (see Table 1).
Prevalence of school alienation and aggressive behaviors at school Descriptive statistics of the outcome measures of school delinquency are shown in Table 2. With respect to the social withdrawal and alienation of school activities due to students’ fear of being attacked or harmed, 2% (n = 108) of youth reported staying away from online activities, 1.2% (n = 67) avoided activities at school, 0.7% (n = 43) skipped classes at school and 1% reported staying home from school (n = 50). Regarding aggressive behaviors at school, 4.5% (n = 262) reported being in one or more physical fights and 0.7% (n = 42) reported having carried a gun on school property. Several measures of bullying including physical bullying (hitting, pushing, slapping, shoving), coercion (being forced to do something you didn’t want to do) and having property destroyed were tested for their association with the outcome measures of aggressive behaviors at school. The results showed significant relationships existed between fighting on the one hand and physical victimization (chi-square = 493.016, p < 0.000), coercion (chi-square = 58.640, p < 0.000), and having property destroyed (chi-square = 204.186, p < 0.000) on the other. As well, gun-carrying was shown to be significantly associated with both physical victimization (chi-square = 8.047, p = 0.005) and coercion (chi-square = 5.816, p = 0.016).
116
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
Fig. 2. Conditional response probabilities.
Subtypes of cyber- and traditional-victim classes Latent class analysis was conducted on the 11 victimization variables with one, two, three, four and five classes specified. The model fit statistics are reported in Table 3, which shows how the AIC, BIC, a-BIC and entropy compared across models. The best fitting model was deemed to be the four-class model based on the BIC, a-BIC, and entropy values. In addition, the Lo–Mendell–Rubin adjusted LRT-test and the Vuong–Lo–Mendell test yielded a p-value of 0.087 and 0.086 (not shown), respectively. Accordingly, the 2, 3 and 4 class models each fit the data better than the next highest class. Therefore, all of the fit statistics indicated that 5-class model did not fit the data better than the 4-class model. In addition, the 4-class model provided a more interpretable solution for analysis. Therefore, the 4-class model was deemed to provide the best fit to these data. The class prevalence and conditional item response probabilities for the four-class model are shown in Table 4. The item response probabilities are plotted in Fig. 2. For each class, the item response probability shows the probability that an individual was victimized by the specific type of behavior. The four latent class model yielded an interpretable solution in terms of the division of victimization classes into the following categories (see Table 3). One class represented youth who fell victim to multiple forms and types of bullying and cyberbullying behaviors (Class 4). The conditional probability of being victimized was high across the majority of measures including being threatened both in person (0.695) and via text message (0.508), being physically and (0.811) verbally bullied (0.979), having rumors spread about them (1.00) and having hurtful information posted about them over the Internet (0.455); a second class of youth who had very low probabilities of being cyberbullied (all probabilities were 0.05 or less) but high probabilities of being verbally abused (0.758), including being the object of rumors (0.653), and physically abused (0.397) (Class 2); a third class of youth who similar to Class 2, were abused verbally (Class 3). The distinguishing feature of class 3, however, was the relatively higher likelihood of being cyberbullied: class 3 youth had higher conditional probabilities of Table 4 Item response probabilities and class prevalence of the 4-class LCM of adolescent victimization.
Verbal Rumor spreading Threatened w/harm Physical Coercion Social exclusion Property destroyed Hurtful information posted Private information shared Threatened online (including gaming) Threatened via text Online exclusion Class prevalence
Non-victimized
Relational/verbal/physical victim
Relational/verbal/cyber victim
Highly victimized
0.043 0.049 0.001 0.001 0.007 0.003 0.005 0.006 0.000 0.002 0.004 0.002 77.3% (4,353)
0.758 0.653 0.205 0.397 0.125 0.213 0.103 0.018 0.000 0.000 0.050 0.014 8.0% (449)
0.526 0.855 0.182 0.054 0.061 0.240 0.032 0.406 0.117 0.195 0.482 0.091 11.6% (653)
0.979 1.000 0.695 0.811 0.356 0.585 0.381 0.455 0.208 0.373 0.508 0.158 3.1% (177)
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
117
Table 5 Parameter estimates, Z-scores, and p-values for the multinomial LC regression of sociodemographic characteristics on class membership with nonvictimization as the reference class. Logit/OR Relational/verbal/cyber v non-victim 1.326/3.77 Female 0.088/1.09 Age Grade −0.013/0.987 White 0.375/1.45 Relational/verbal/physical vs. non-victim −0.501/0.606 Female Age −0.136/0.873 Grade −0.216/0.801 White 0.029/1.03 Highly victimized vs. non-victim Female 1.321/3.75 Age −0.136/0.873 Grade 0.108/1.11 White −2.850/0.058 Intercepts C#1 11.523 C#2 22.913 C#3 27.457
Z-score
p-Value
5.107 1.225 −0.151 1.937
Contents lists available at ScienceDirect
Child Abuse & Neglect
The association between school exclusion, delinquency and subtypes of cyber- and F2F-victimizations: Identifying and predicting risk profiles and subtypes using latent class analysis Gia Elise Barboza ∗ African American Studies and Criminology, Northeastern University, 200F Renaissance Park, 360 Huntington Ave., Boston, MA 02115, USA
a r t i c l e
i n f o
Article history: Received 1 March 2014 Received in revised form 6 August 2014 Accepted 8 August 2014 Available online 5 September 2014 Keywords: Cybervictimization Bullying Academic performance Aggression Latent class analysis
a b s t r a c t This purpose of this paper is to identify risk profiles of youth who are victimized by onand offline harassment and to explore the consequences of victimization on school outcomes. Latent class analysis is used to explore the overlap and co-occurrence of different clusters of victims and to examine the relationship between class membership and school exclusion and delinquency. Participants were a random sample of youth between the ages of 12 and 18 selected for inclusion to participate in the 2011 National Crime Victimization Survey: School Supplement. The latent class analysis resulted in four categories of victims: approximately 3.1% of students were highly victimized by both bullying and cyberbullying behaviors; 11.6% of youth were classified as being victims of relational bullying, verbal bullying and cyberbullying; a third class of students were victims of relational bullying, verbal bullying and physical bullying but were not cyberbullied (8%); the fourth and final class, characteristic of the majority of students (77.3%), was comprised of non-victims. The inclusion of covariates to the latent class model indicated that gender, grade and race were significant predictors of at least one of the four victim classes. School delinquency measures were included as distal outcomes to test for both overall and pairwise associations between classes. With one exception, the results were indicative of a significant relationship between school delinquency and the victim subtypes. Implications for these findings are discussed. © 2014 Elsevier Ltd. All rights reserved.
Introduction The upsurge in school shootings during the past decade has been paralleled by a dramatic increase in research on the criminogenic effects of peer victimization at school (Juvonen & Graham, 2001; Garbarino & DeLara, 2004; Rubinlicht, 2011; Hong & Espelage, 2012). The impact that bullying has on future criminal behavior and/or delinquency has made it a particularly fruitful area of investigation. While bullying typically conjures up images of physical intimidation, the reality is that youth are victimized in a variety of ways, any or all of which may result in irreparable psychological and/or physical harm (Cullen, Unnever, Hartman, Turner, & Agnew, 2008). In addition to overt acts of hostility, covert forms of harassment
∗ Corresponding author. http://dx.doi.org/10.1016/j.chiabu.2014.08.007 0145-2134/© 2014 Elsevier Ltd. All rights reserved.
110
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
including spreading rumors or lies and excluding and/or ignoring other students from school activities are commonplace. Moreover, as a result of today’s electronic age, the Internet is supplanting traditional forms of schoolyard bullying however victims characteristics differ according to the type of harm being inflicted (Law, Shapka, Hymel, Olson, & Waterhouse, 2012). The differences in form and function between traditional bullying and cyberbullying begs for consideration of whether this “new” form of harassment (Wade & Beran, 2011) that uses “email, instant messages, cell phones, text messages, photos, videos and social networking websites to humiliate and threaten others” (Grim, 2008, p. 157) is really just a new way to implement the same behavior or rather whether it is indeed a distinct phenomenon. Despite significant advances in our understanding of bully victimization over the past decade, much remains to be understood. An important area of research is to examine potential overlaps in harassment and bullying across all possible environments to gain understanding of the sum total of youth’s experiences with victimization (Ybarra, Diener-West, & Leaf, 2007). For example, what victim profiles emerge when multiple indicators of both online and offline harassment are analyzed? Also, what is the association between victim profiles and delinquent behavior at school such as skipping class, absenteeism and/or weapon-carrying and physical fighting? The aim of this paper is to document the co-occurrence of online and face-to-face peer harassment and bullying and its effect on several measures of school alienation, avoidance and delinquency. Latent class analysis is used to identify the nature and form of victimization and bullying in early adolescence, explore the overlap and co-occurrence among different clusters of victims, and to examine the relationship between multiple risk factors for school exclusion, delinquency and membership in each ‘victim’ class. Prevalence of bullying and cyberbullying Aggression includes any single act of physical, emotional and/or psychological harm including physical or sexual violence, psychological or emotional abuse or neglect (U.S. Department of Justice, 2004). Common to most definitions of bullying, on the other hand, is that it is repetitive, results in harm and occurs in the context of an imbalance of either psychological or physical power. While fights, one-time attacks, or “harmless” teasing are appropriately considered aggressive behaviors (Berger, 2007), individuals subjected to these same behaviors repeatedly are considered bullied in a more general sense. Examples of bullying include being called names, being physically hurt, being threatened, being the subject of rumors, being socially isolated and having one’s belongings taken repeatedly (Glew, Rivara, & Feudtner, 2000). Similar to traditional bullying, cyber bullying is often defined as “willful and repeated harm” (Hinduja & Patchin, 2010, p. 5) inflicted toward another. In contradistinction to traditional face-to-face bullying, however, cyberbullying utilizes electronic communication to threaten, harass embarrass, or socially exclude (Hinduja & Patchin, 2010; Patchin & Hinduja, 2006; Williams & Guerra, 2007). While there is no universally accepted definition of cyberbullying (Smith, del Barrio, & Tokunaga, 2013), the very nature of the Internet as being anonymous, accessible and wide-reaching implies that all the elements of bullying are present in a single online interaction. Cyberbullying involves repetition not only because material such as email, text, or pictures can be viewed by anyone with online access (Campbell, 2005; Slonje & Smith, 2008) but also because the same communication can be repeated a number of times through mass communication. The anonymous nature of the interaction endows the perpetrator with power regardless of his or her personal characteristics. The difficulty associated with removing unwanted or harmful communication as well as its permanency only serves to reinforce the power and repetitiveness of the interaction (Wolak, Mitchell, & Finkelhor, 2007). A large body of research documenting the prevalence, magnitude and impact of peer-to-peer victimization challenges commonly held assumptions that school is a place where children are safe. Nansel, Overpeck, and Pilla (2001) at the National Institute of Child Health reported that bullying affects approximately 30% of adolescents in the United States, 13.6% as bullies, 10.6% as victims, and 6.3% as bully-victims. While estimates differ greatly due to different methodologies and conceptualizations of “bullying,” most scholars agree that between approximately 15–20% of students are regular victims of bullying behavior while about 10% of children in the US experience “extreme victimization” by bullying (Seals & Young, 2003; Peskin, Tortolero, Markham, Addy, & Baumler, 2007). The most prevalent form of bully victimization is name-calling (13.5%) followed by physical aggression (10.8%), teasing (9.0%), exclusion (7.3%) and threats of harm (6.6%) (Seals & Young, 2003). Characteristics of bully and cyberbully victims Initial research on bullying reported that physical bullying declines as youth get older (Perry, Kusel, & Perry, 1988), however later research clarified the existence of a parabolic relationship between physical bullying and age. Bullying is less common among younger and older students, peaking sometime during elementary school (Borg, 1998) around age 7, and declines steadily thereafter through the age of 15 (Glew et al., 2000). Although physical bullying tends to lessen over time, verbal abuse continues to remain high and to be associated with physical abuse as children age (Perry et al., 1988) Type of bully victimization differs according to gender. Girls are more often the targets and perpetrators of passive, indirect bullying, also known as “relational” bullying, such as being the object of gossip and social isolation. Boys on the other hand are more commonly the targets and perpetrators of aggressive, physical bullying (Glew et al., 2000). More recent studies confirm that boys experience physical bullying to a greater extent than girls, the latter being more likely to report being teased or joked about (Carbone-Lopez, Esbensen, & Brick, 2010). Similarly, Dukes, Stein, and Zane (2010) found that adolescent girls reported more relational victimization and adolescent boys reported more physical bullying and victimization, more weapon carrying, and more injury.
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
111
Consistent with the literature on victimization, some researchers have found that the prevalence of victimization by bullying is higher among racial and ethnic minority youth, but results are inconsistent (Peskin et al., 2007) and little is known about racial/ethnic differences in bullying risk (Spriggs, Iannotti, Nansel, & Haynie, 2007). Seals and Young (2003) found no significant differences in the prevalence of victimization between African American and white students. More recently, researchers have documented racial/ethnic differences in bully prevalence, with black students reporting less victimization than both whites and Hispanics (Spriggs et al., 2007). On the other hand, Nansel et al. (2001), who used a representative national sample of youth in grades 6 through 10, found that black youth reported being bullied significantly more than either whites or Hispanics. In yet a different study, Hispanic students reported less victimization than white or African American students however, when they were victimized, they were more likely to experience repeat victimization (Hanish and Guerra, 2002). Clearly, more work needs to be done in order to partial out the effects of race and ethnicity on bully victimization. Schools as correlates and consequences of bullying and cyberbullying Existing research has consistently shown that the interpersonal and institutional settings in which student interactions take place play an important role in the psychosocial functioning of students. Researchers conceptualize bullying (Craig & Pepler, 2007) and cyberbullying (Hoff and Mitchell, 2009; Mishna, Saini, & Solomon, 2009) as part of the broader context of embedded socio-ecological relationships. Accordingly, theorizing about bullying behaviors in children and adolescents implicates both individual and contextual factors (Barboza et al., 2009). The school environment, in particular, is a place where students’ beliefs about violence are formed and where students learn to model the behavior of their peers and adults (Osterman, 2000). Therefore, schools are key in helping us understand the complexity of bullying and for developing sensitive and effective interventions (Barboza et al., 2009; Limber, 2006; DeLara, 2006; Garbarino & DeLara, 2004; Bronfenbrenner and Morris, 1998). Bullying, cyberbullying, internalizing behavior at school Children who are victimized are more likely to report increased levels of internalizing symptoms, such as depression, suicidal ideation and loneliness, not only in middle (Seals & Young, 2003) and high school (Peskin et al., 2007) but for years afterwards. Early research focused on the effect of school size, class size and competition at school on subsequent bullying but attention later shifted to explore different aspects of the school environment such as alienation and school-related stress. Previous research has shown that a wide variety of school characteristics are associated with the presence of both bullying and cyberbullying. These include school-related stressors (e.g. unreasonable expectations of student performance), teacher apathy and/or lack of support, school atmosphere (e.g. feelings of belonging, fair rules) and the broader peer environment including both the presence of peers (Atlas & Pepler, 1998; Craig & Pepler, 2007) and the nature of peer-to-peer relationships (Barboza et al., 2009). One investigation found that positive disciplinary actions and high levels of adult supervision, school involvement and academic achievement characterize schools reporting low levels of bullying. Using grade point average as an objective measure of academic achievement, Glew, Fan, Katon, and Rivara (2008) reported that victims have significantly lower academic achievement in school than non-victims. In sum, research supports the claim that victims have lower levels of affective attachment to school: they are neither as involved nor as committed to school compared to non-victims (Spriggs et al., 2007). In comparison to non-victimized youth, cyberbullied youth report more sadness, anxiety and fear (Beran & Li, 2008); have difficulty with social interactions (Ybarra, Mitchell, Wolak, & Finkelhor, 2006; Blais, 2008); are more likely to be engaged in risky behavior such as drug and alcohol use (Mishna, Cook, Gadalla, Daciuk, & Solomon, 2010) and eating disorders (DeHue, Bolman, & Völlink, 2008; Fosse & Holen, 2006). On this basis, social and emotional functioning is likely to have a mediating impact on the relationship between bullying and educational outcomes. According to one study, 7% of America’s 8th graders stay home at least one day a month because they are afraid of other children at school. Fear of school, in turn, is associated with lower self-esteem and poor school performance. Research has shown that after adjusting for age, sex, ethnicity and income, victims of traditional bullying are significantly more likely than bystanders and non-victims to feel unsafe at school and more likely to feel like they “do not belong” at school (Glew et al., 2008). Students who are bullied in cyberspace only, and students bullied both in cyberspace and at school, experience difficulties at school such as low marks, poor concentration and absenteeism (Beran & Li, 2008). Moreover, they are less able to concentrate at school (Beran & Li, 2005; Juvonen & Gross, 2008), which most likely results in the greater likelihood of skipping school (Ybarra et al., 2007), and receiving detentions and suspensions (Wolak et al., 2007; Ybarra et al., 2007) among victims. Differences in frequency and type of victimization may be important determining factors for predicting compromised psychosocial ability and academic performance characteristic of abused children. Bullying, cyberbullying and externalizing behavior at school. There is a clear relationship between being the victim of bullying and subsequent aggression. As scholars have begun to explore the link between bully victimization and aggressive behavior (Anderson et al., 2001), many have found that a subset of bullied youth respond aggressively to their victimization while others respond more “passively.” Passive victims have greater feelings of insecurity, a negative worldview and are more like to view themselves as a failure in comparison to non-victims (Glew et al., 2000). Provocative victims, on the other hand,
112
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
are not afraid of confrontation. As a result, provocative victims are at greater risk of engaging in violent behavior. Research has shown that bully-victim relationships underlie the majority of childhood assaults, suicides and homicides with bullyvictims at risk for committing suicide and/or murder. Anderson et al. (2001) studied school-associated violent deaths in the US between 1994 and 1999 and found that homicide perpetrators are more than twice as likely as homicide victims to have been victims of bullying at school. Several additional studies since this seminal piece have confirmed the relationship between weapon-carrying and bully victimization. For example, youth who reported being targeted by Internet harassment were eight times more likely than all other youth to concurrently report carrying a weapon to school (Ybarra et al., 2007). An analysis of 15 school shootings the occurred between 1995 and 2001 revealed that in 87% of the cases, shooters had experienced relational bullying, especially chronic rejection (Leary, Kowalski, Smith, & Phillips, 2003, Dukes et al., 2010). Using data from 1,300 adolescents in a Colorado school district, Dukes et al. (2010) found that being a victim of physical bullying, a victim of relational bullying, or being a relational bully significantly predicted more injury among both boys and girls. The devastation that results from the confluence of victimization by bullying, weapon-carrying and social withdrawal is permanently established in the public’s consciousness due to recent school massacres occurring at elementary, high schools and universities across the country. Bully and cyberbully typologies and associated risk profiles The terms “bullying” and “cyberbullying” encompass a variety of different behaviors and experiences (Cornell & Mehta, 2011; Lovegrove & Cornell, 2013) that may or may not be independent of each other (Wang, Iannotti, Luk, & Nansel, 2010). Bully victimization is multifaceted and includes verbal, physical and/or psychological aggression. An emergent body of research documenting the overlap and co-occurrence of these behaviors exemplifies their heterogeneity. For example, although physical and verbal harassment are conceptually distinct, empirically victims are more likely to be verbally harassed given that they have experienced physical abuse. Prior research has shown that as many as 58% of students report being victims of both verbal and physical bullying (Orpinas, Horne, & Staniszewski, 2003). In addition, the experiences of frequently victimized children fall along distinct dimensions (Wang et al., 2010) each characterized by different coping strategies and adjustment problems (Waasdorp & Bradshaw, 2011). For example, researchers have found distinct classes of victims including an ‘all-types’ victim class (9.7% of males and 6.2% of females), a verbal/relational victims class (28.1% of males and 35.1% of females), and a non-victim class (62.2% of males and 58.7% of females) (Wang et al., 2010). A recent study conducted by Lovegrove and Cornell (2013) used latent class analysis to study the association between bullying involvement and internalizing and externalizing problem behaviors among over 3,500 high school students from Virginia. They found four latent classes of bullies and victims composed of a non-involved class (65%), a bullies class (12%), a victims class (16%), and a bully-victims class (8%). Externalizing problem behaviors were significantly higher among students in the bullies and bully-victims classes, while internalizing problem behaviors were higher among victims and bully-victims. These analyses directly or indirectly point to the fact that both polyvictimization and the frequency of abuse has more adverse effects on internalizing and externalizing symptoms such as psychological well-being, academic achievement and criminogenic effects (Berger, 2007; Buhs, Ladd, & Herald, 2006; Hawker & Boulton, 2000; Nansel et al., 2001). The current study The present study expands upon previous work in several important ways. First, unlike previous studies that model bully victimization as a random and independent event, the present analysis models the interdependent nature of victimization by allowing for the possibility that more than one form of victimization is part of the bully-victim experience. Secondly, it is one of the first studies to utilize latent class analysis to explore the co-occurrence of cyberbullying and bullying victimizations in a nationally representative sample using five separate indicators of both types of behavior (10 all together) (but see Wang et al., 2010). In the analysis that follows, conditional class means are used to describe the latent structure of bullying – both traditional and online – among youth. The current study also addresses the question of which victimhood classe(s) impacts school participation and delinquency, including weapon carrying and fighting, the most—a question not previously addressed in the existing literature. To do so, only academic outcomes that have directly resulted from students’ own experiences with victimization are utilized. Analytically, the latent classes are used as predictors of school outcomes and are tested for significant associations. This allows for the possibility of quantifying the impact of victimization across several academic indicators, providing vital information regarding potential targets for the prevention of academic and behavior problems in children. More specifically, this paper addresses the following research questions: 1. What is the prevalence and co-occurrence of online and face-to-face victimization by bullying; 2. How can we describe the risk profiles of victims? Do victims of bullying and cyberbullying have distinct profiles (i.e. is there a separate class of cyberbullied and bullied) or are they both part of the same phenomenon (i.e. are victims of bullying likely to be cyberbullied and vice versa). In order to address this question, a latent class analysis of victims was implemented to predict class prevalence as well as the conditional probability of being a victim latent class membership; 3. How do profiles of risk vary with victims’ sociodemographic characteristics? A multinomial logistic latent class regression was performed using characteristics such as age, gender, race, and grade to predict latent class membership.
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
113
4. Does membership in a victim class predict school behavior? Posterior probabilities were used to predict latent class membership and then pairwise comparisons were performed for the purpose of testing the equality of means across 6 measures of school behavior (including skipping classes, being absent from school, fighting at school and carrying a weapon to school). Methods Participants Data for this study were derived from the 2011 National Crime Victimization Survey: School Crime Supplement (SCS). Twice each year, data are obtained from a nationally representative sample of roughly 49,000 households comprising about 100,000 persons on the frequency, characteristics, and consequences of criminal victimization in the United States. Youth between 12 and 18 years old living in a household selected to complete the NCVS are eligible for inclusion in the SCS. The final sample includes persons between 12 and 18 years of age who were in primary or secondary education programs leading to a high school diploma (elementary through high school), and who had been enrolled sometime during the 6 months prior to the interview. The SCS asks questions related to students’ experiences with, and perceptions of, crime and safety at school, student bullying, and fear of victimization at school. Mplus 6.12 (Muthen & Muthen, 1998–2010) was the software used to analyze data for the current study.
Measures Demographic variables Demographic variables included age, gender, grade (5th–12th) and race which was coded as white/non-white irrespective of ethnicity.
Peer victimization Seven items from the SCS were utilized to measure students’ experiences with bully victimization. The participants were asked the following, “Now I have some questions about what students do at school that make you feel bad or are hurtful to you. We often refer to this as being bullied. You may include events you told me about already. During this school year, has any student bullied you?” The responses were coded as either direct or indirect bullying. In accordance with previous literature, direct bullying refers to situations in which the victim is directly involved whereas indirect bullying may take place without the victims immediate knowledge or presence (Vandebosch & Van Cleemput, 2009). Therefore, direct bullying included physical bullying (“Pushed you, shoved you, tripped you, or spit on you?”), verbal bullying (“made fun of you, called you names, or insulted you, in a hurtful way”), being threatened with harm, being coerced into doing things you don’t want to do, being excluded from activities on purpose and having property destroyed. Indirect bullying was measured with one item, namely tapping whether the respondent was the object of rumors or other forms of communication intended to make others dislike them (“spreading rumors about you or tried to make others dislike you”). Students’ experiences with cyberbullying were measured with four items that were similarly classified as either direct or indirect. Direct cyberbullying included threats of harm made online (“whether another student used the internet to threaten or insult you”), online exclusion (“being purposefully excluded from online communications”), and non-verbal harm (“someone posted hurtful information about you on the Internet”), while indirect bullying was measured as outing (“purposefully sharing your private information, photos or video”).
School alienation Four items were used to measure school withdrawal/avoidance. Participants were asked the following questions: (1) “during this school year, did you STAY AWAY from any online activities because you thought someone might be mean to you there?” (2) “Did you AVOID any activities at your school because you thought someone might attack or harm you?” (3) “Did you AVOID any classes because you thought someone might attack or harm you?” and (4) “Did you stay home from school because you thought someone might attack or harm you in the school building, on school property, on a school bus, or going to or from school?”
Aggressive behaviors at school. Two items measured aggressive behavior at school. Students were asked, “During this school year, have you been in one or more physical fights at school?” and During this school year, did YOU ever bring the following to school or onto school grounds?” “A gun?” All response choices were binary (1 = Yes/2 = No).
114
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
Fig. 1. A latent class mixture model of traditional and cyberbullying with academic risk as a distal outcome and youth demographic characteristics as predictors of class membership.
Statistical analyses Statistical analyses were conducted using MPLUS 6.2. First, descriptive statistics of all variables were examined. Second, a series of latent class models were estimated (1) to explain the relationships among a set of observed measures of victimization by means of discrete but latent groups; (2) to identify “types” or classes of victims; and (3) to derive an estimate of the prevalence of each class. The first step in LCA Competing solutions were compared on the basis of model fit indices, interpretability, and parsimony. Determination of model fit was based on log-likelihood, Akaike Information Criteria (AIC), Bayes Information Criteria (BIC), sample size adjusted Bayes Information Criteria (a-BIC), entropy, and Lo–Mendel–Rubin adjusted likelihood ratio test (LMR) (Lo et al., 2001). Smaller values of log-likelihood, AIC, BIC and a-BIC indicate better fit to the data whereas significant p-values associated with an LMR indicate significant improvement in model fit relative to the solution with one less class. Once a final solution was identified, the model was adjusted for age, gender, race and grade level. After the final model was estimated, auxiliary variables measuring school alienation/withdrawal and aggressive behavior were used as outcome variables. This approach removed the auxiliary variables’ influence over the latent structure and hence did not change the substantive interpretation of the classes (Wang et al., 2010). The analytical model is represented in Fig. 1.
Results Sample characteristics The sample consists of 5,589 youth between the ages of 12 and 18 (mean = 14.77, s.d. = 1.99). Their highest level of academic achievement was the 9th grade, on average. Males and females comprised 51% and 49% of the sample, respectively. The majority of respondents identified racially as “white” (80%) and the rest as non-white (20%) including African/American or black, Hispanic and/or Asian.
Descriptive characteristics Prevalence of cyber and bully victimization. Descriptive statistics of the bullying behaviors are provided for the whole sample and by gender in Table 1. Table 1 shows the distinction between direct and indirect forms of bullying. The prevalence of victimization from the 7 types of face-to-face bullying were 18.5% for spreading rumors, 17.7% for verbal (males: 17.1, females: 19.7), 7.9 for physical (males: 9.5, females 7.1), 5.5% for social exclusion (males: 4.5, females 6.5), 5.1 for threatened with harm (males: 5.5, females 5.1), 3.3% for being forced to do things (males: 3.5, females: 3.0) and 2.8% for destroyed property (males: 3.5, females: 2.2). Being threatened or insulted online, via text messaging or via gaming (i.e. verbal cyberbullying) was the most common form of cyberbullying (7.0%; n = 237) followed by non-verbal cyberbullying (subjected to the posting of hurtful information; 3.7%; n = 201), social exclusion (purposefully being excluded from online communications; 1.2%; n = 64) and then outing (having private information, photos or videos on the internet or via phone shared in a hurtful way (1.1%; n = 58). The single measure reported here of being threatened is a constellation of three variables: online threats, threats received from text messaging and on threats received via gaming websites. In the analyses that follow, being threatened by text message was kept as a separate variable but gaming and online threats were consolidated. Consequently, there are 11 victimization variables in Table 1 but 12 victimization variables reported in Table 3.
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
115
Table 1 Prevalence of bullying and cyberbullying behaviors and chi-square tests of association by gender 1. Prevalence of traditional and cyber victimizations among peers by gender. Traditional Victimization 1. Direct Verbal (i.e. calling someone names) Threatened w/harm Physical (i.e. hitting, pushing) Forced to do things Social (i.e. excluding someone from group activities) Property (i.e. destroying someone’s property) 2. Indirect Spread false rumors Cybervictimization 1. Direct Verbal (i.e. using internet to threaten or insult someone) Social (i.e. excluding someone from online communications) Non-verbal (i.e. posting hurtful information) 2. Indirect Outing (i.e. sharing personal information)
Males
Females
Total
Odds ratio
17.1 5.2 9.5 3.5 4.5 3.5
19.7 5.3 7.1 3.0 6.5 2.2
17.7 5.1 7.9 3.3 5.5 2.8
1.19*** 1.02 0.728*** 0.853 1.48*** 0.620***
12.8
24.5
18.5
2.21***
5.4 0.9 1.8
8.8 1.5 5.9
7.0 1.2 3.7
1.69*** 1.68** 3.42***
0.7
1.5
1.1
2.16***
Table 2 Outcome measures: school withdrawal and avoidance due to fear of being attacked, and aggressive behaviors at school. Social withdrawal from/avoidance of school
n (%)
Stay away from online activities Avoid activities at school Skip classes at school Stay home from school Aggressive behaviors at school Been in one or more physical fights Carried a gun to school
108 (2.0) 67 (1.2) 43 (0.7) 50 (1.0) 262 (4.5) 42 (0.7)
Table 3 Model Fit Statistics for the n-class models of peer victimization. Akaike (AIC)
Bayesian (BIC)
Sample size adjusted BIC (a-BIC)
Entropy
Lo, Mendell, Rubin test
n for each class
1-LC 2-LC 3-LC 4-LC
28,693.177 22,786.01 22,396.149 22,209.327
28,773.16 22,952.641 22,649.429 22,549.256
28,735.027 22,873.199 22,528.676 22,387.193
na 0.899 0.883 0.885
na 5,871.567 412.201 210.792
5-LC
22,127.912
22,554.489
22,351.116
0.859
106.393
C1 = 5,798 C1 = 4,770; C2 = 1,028 C1 = 4,773; C2 = 816; C3 = 209 C1 = 4,763; C2 = 662; C3 = 242; C4 = 131 C1 = 4,507; C2 = 741; C3 = 284; C4 = 163; C5 = 103
The results show that bully victimizations differ significantly by gender. The odds of being bullied by rumors or being subjected to the unwanted posting of hurtful information, for example, are 2.11 (p < 0.000) and 3.58 (p < 0.000) times higher for females compared to males (see Table 1).
Prevalence of school alienation and aggressive behaviors at school Descriptive statistics of the outcome measures of school delinquency are shown in Table 2. With respect to the social withdrawal and alienation of school activities due to students’ fear of being attacked or harmed, 2% (n = 108) of youth reported staying away from online activities, 1.2% (n = 67) avoided activities at school, 0.7% (n = 43) skipped classes at school and 1% reported staying home from school (n = 50). Regarding aggressive behaviors at school, 4.5% (n = 262) reported being in one or more physical fights and 0.7% (n = 42) reported having carried a gun on school property. Several measures of bullying including physical bullying (hitting, pushing, slapping, shoving), coercion (being forced to do something you didn’t want to do) and having property destroyed were tested for their association with the outcome measures of aggressive behaviors at school. The results showed significant relationships existed between fighting on the one hand and physical victimization (chi-square = 493.016, p < 0.000), coercion (chi-square = 58.640, p < 0.000), and having property destroyed (chi-square = 204.186, p < 0.000) on the other. As well, gun-carrying was shown to be significantly associated with both physical victimization (chi-square = 8.047, p = 0.005) and coercion (chi-square = 5.816, p = 0.016).
116
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
Fig. 2. Conditional response probabilities.
Subtypes of cyber- and traditional-victim classes Latent class analysis was conducted on the 11 victimization variables with one, two, three, four and five classes specified. The model fit statistics are reported in Table 3, which shows how the AIC, BIC, a-BIC and entropy compared across models. The best fitting model was deemed to be the four-class model based on the BIC, a-BIC, and entropy values. In addition, the Lo–Mendell–Rubin adjusted LRT-test and the Vuong–Lo–Mendell test yielded a p-value of 0.087 and 0.086 (not shown), respectively. Accordingly, the 2, 3 and 4 class models each fit the data better than the next highest class. Therefore, all of the fit statistics indicated that 5-class model did not fit the data better than the 4-class model. In addition, the 4-class model provided a more interpretable solution for analysis. Therefore, the 4-class model was deemed to provide the best fit to these data. The class prevalence and conditional item response probabilities for the four-class model are shown in Table 4. The item response probabilities are plotted in Fig. 2. For each class, the item response probability shows the probability that an individual was victimized by the specific type of behavior. The four latent class model yielded an interpretable solution in terms of the division of victimization classes into the following categories (see Table 3). One class represented youth who fell victim to multiple forms and types of bullying and cyberbullying behaviors (Class 4). The conditional probability of being victimized was high across the majority of measures including being threatened both in person (0.695) and via text message (0.508), being physically and (0.811) verbally bullied (0.979), having rumors spread about them (1.00) and having hurtful information posted about them over the Internet (0.455); a second class of youth who had very low probabilities of being cyberbullied (all probabilities were 0.05 or less) but high probabilities of being verbally abused (0.758), including being the object of rumors (0.653), and physically abused (0.397) (Class 2); a third class of youth who similar to Class 2, were abused verbally (Class 3). The distinguishing feature of class 3, however, was the relatively higher likelihood of being cyberbullied: class 3 youth had higher conditional probabilities of Table 4 Item response probabilities and class prevalence of the 4-class LCM of adolescent victimization.
Verbal Rumor spreading Threatened w/harm Physical Coercion Social exclusion Property destroyed Hurtful information posted Private information shared Threatened online (including gaming) Threatened via text Online exclusion Class prevalence
Non-victimized
Relational/verbal/physical victim
Relational/verbal/cyber victim
Highly victimized
0.043 0.049 0.001 0.001 0.007 0.003 0.005 0.006 0.000 0.002 0.004 0.002 77.3% (4,353)
0.758 0.653 0.205 0.397 0.125 0.213 0.103 0.018 0.000 0.000 0.050 0.014 8.0% (449)
0.526 0.855 0.182 0.054 0.061 0.240 0.032 0.406 0.117 0.195 0.482 0.091 11.6% (653)
0.979 1.000 0.695 0.811 0.356 0.585 0.381 0.455 0.208 0.373 0.508 0.158 3.1% (177)
G.E. Barboza / Child Abuse & Neglect 39 (2015) 109–122
117
Table 5 Parameter estimates, Z-scores, and p-values for the multinomial LC regression of sociodemographic characteristics on class membership with nonvictimization as the reference class. Logit/OR Relational/verbal/cyber v non-victim 1.326/3.77 Female 0.088/1.09 Age Grade −0.013/0.987 White 0.375/1.45 Relational/verbal/physical vs. non-victim −0.501/0.606 Female Age −0.136/0.873 Grade −0.216/0.801 White 0.029/1.03 Highly victimized vs. non-victim Female 1.321/3.75 Age −0.136/0.873 Grade 0.108/1.11 White −2.850/0.058 Intercepts C#1 11.523 C#2 22.913 C#3 27.457
Z-score
p-Value
5.107 1.225 −0.151 1.937
