Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
An examination of in-school and online protective factors for adolescent trajectories of online victimization and harassment over time
(USC Thesis Other)
An examination of in-school and online protective factors for adolescent trajectories of online victimization and harassment over time
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
An Examination of In-School and Online Protective Factors for Adolescent Trajectories of Online Victimization and Harassment Over Time by Sophia Caitlin Hiss A Dissertation Presented to the faculty of the USC Graduate School University of Southern California in partial fulfillment of the requirements for the degree Doctor of Philosophy In Education December 2015 ii Acknowledgements As with anything in life, I could not have completed this PhD program or dissertation without the help and support of a number of people. First, I would be nowhere without the love and support of my parents and brother. Thank you for supporting me in all of my life plans, even if they did not make sense at the time. A very big thank you goes to my partner Sarah. Your love and support have gotten me through one of the biggest challenges in my life. I would also like to thank my advisor Brendesha Tynes. I could not have finished this without your support and guidance through both the dissertation and PhD program. I would also like to thank the other members of my committee Drs. Jack McArdle and Robert Rueda. I have learned so much from each of you and will be forever indebted to your kindness and support as I moved through this PhD program. I would also like to thank my cohort. Though it both seems like an eternity and no time at all, we got through this. Finally, I would like to thank my friends in and out of the PhD program. Thank you all for listening to me vent and rant as well as joining in on the celebrations. To the Ktown study group, I know there is no better way to spend a whole day at a coffee shop. iii TABLE OF CONTENTS Acknowledgements ......................................................................................................................... ii LIST OF TABLES .......................................................................................................................... v LIST OF FIGURES ....................................................................................................................... vi ABSTRACT .................................................................................................................................. vii CHAPTER 1: INTRODUCTION AND THEORETICAL FRAMEWORK .................................. 1 Purpose of Study ......................................................................................................................... 2 Definitions ................................................................................................................................... 2 Closer Examination of Online Victimization and Cyberbullying ........................................... 4 Theoretical Framework ............................................................................................................... 7 Social Cognitive Theory .......................................................................................................... 8 Risk and Resilience Theory ................................................................................................... 10 CHAPTER 2: LITERATURE REVIEW ...................................................................................... 19 Consequences of Involvement in Online Victimization ........................................................... 19 Transitions among Harasser and Victim Groups Over Time ................................................ 22 Methodological Issues ............................................................................................................... 23 CHAPTER 3: METHOD .............................................................................................................. 27 Data ........................................................................................................................................... 28 Project Description ................................................................................................................ 28 Victimization and Harassment Variables .............................................................................. 31 Social Support Variables ....................................................................................................... 34 Analyses................................................................................................................................. 37 CHAPTER 4: RESULTS .............................................................................................................. 39 Descriptive Statistics ................................................................................................................. 39 Model Building ......................................................................................................................... 40 Question 1: Are there distinct groups of adolescent online victimization and harassment behavior? Do adolescents transition between groups over time? .......................................... 43 Question 2: Does social support from teachers, peers (online and offline), or family predict initial patterns in online victimization participation? ............................................................ 46 Question 3: Does social support from teachers, peers (online and offline), or family predict transitions between statuses over time? ................................................................................. 48 iv CHAPTER 5: DISCUSSION ........................................................................................................ 53 Summary of Findings ................................................................................................................ 53 Theoretical Implications of Findings ........................................................................................ 58 Social cognitive theory .......................................................................................................... 59 Risk and resilience ................................................................................................................. 60 Limitations ................................................................................................................................ 62 Future Directions for Research ................................................................................................. 66 Contributions to the Literature .................................................................................................. 67 Implications for Schools............................................................................................................ 70 Conclusion ................................................................................................................................. 71 References ......................................................................................................................................72 Tables .............................................................................................................................................93 Figures..........................................................................................................................................102 v LIST OF TABLES Table 3.1 Items Used in LTA and LCA Models ....................................................................93 Table 4.1 Means and Standard Deviations For All Variables................................................94 Table 4.2 Correlation Matrix for All Variables .....................................................................95 Table 4.3 Fit indices for LCA models with 3–5 Classes for Each Wave ..............................96 Table 4.4 Items Response Probabilities and Status Prevalences at Waves One and Two .....97 Table 4.5 Transition Probabilities Table ................................................................................98 Table 4.6 Logistic Regression Results For Covariates Predicting Wave One Status ............99 Table 4.7 Count of Participants for Each Transition ...........................................................100 Table 4.8 Odds Ratios for Social Support Variables Predicting Transitions from Wave One Statuses to Wave Two Statuses ....................................................................101 vi LIST OF FIGURES Figure 3.1. Line graph showing relative error by the number of splits ......................102 Figure 3.2 Final DTA model for attrition ..................................................................103 Figure 4.1 LCA Probabilities at Wave 1 LCA models with 3–5 Classes for Each Wave ...............................................................................................104 vii ABSTRACT Research has found that online victimization and harassment are related to negative outcomes for both victims and harassers. Further, a group involved in both victimization and harassment, harasser-victims, has the most detrimental outcomes. Using risk and resilience and social cognitive theories, I examined the risk for, development of, and factors related to cessation of online harassment and/or victimization over two waves of data. I employed latent transition analysis to examine underlying or latent groups, called statues, in the data. Latent transition analysis also allowed me to examine how adolescents changes status membership over time. By adding in covariates I investigated the relationship between social support from peers, family, and teachers and both initial statuses as well as transitions between statuses. Results indicated that offline peer, family, and teacher support were each related to higher odds of membership in the uninvolved class than the harasser-only, victim-only, or harasser-victim classes. Online peer support, however, was related to lower odds of membership in the uninvolved status than any of the other three statuses. Examination of the relationship between covariates and the transitions indicated that higher levels of offline peer, online peer, and family support were protective- stabilizing for some participants in the harasser-only status at wave one. However, all four types of social support were also related to greater odds of transitioning to the harasser-victim and victim-only statuses for other participants in the harasser-only status. Further, higher levels of teacher support and online peer support were related to higher odds of uninvolved participants transitioning to the victim-only status. Whereas family and offline peer support were related to lower odds of transitioning from uninvolved to the harasser-victim status at wave two. These results seem confounding, but the differences may be due to the type or quality of support an adolescent received. Though the model is exploratory in nature, the results indicate that schools viii and teachers as well friends and family members may have some influence over the changes in online harassment and victimization for adolescents. 1 CHAPTER 1: INTRODUCTION AND THEORETICAL FRAMEWORK A majority of young people have access to the internet, as evidenced by a Pew Research Center (2013) study which found that 95% of all teenagers use the internet, 93% have access to a computer at home, and many report access to the internet at local libraries or cybercafés. Once online, 80% of teenagers report using social networking sites such as Facebook, Twitter, or Instagram, 88% of those users report witnessing harassment of someone online (Lenhart et al., 2011), and somewhere between 10-40% of adolescents have been harassed by others online (Kowalski, Giumetti, Schroeder, & Lattanner, 2014). Further, research examining outcomes related to these online experiences have found a relationship between victimization and poor psychological health, emotional distress, and negative school behaviors (Bauman, Toomey, & Walker, 2013; Klomek, Sourander, & Gould, 2010; Li, 2007; Livingstone & Smith, 2014; Navarro, Yubero, Larrañaga, & Martínez, 2012; Patchin & Hinduja, 2010). However, these negative outcomes are not solely experienced by victims of online harassment. Those who harass others or experience both victimization and harassment also report psychological and social problems and research shows that this group suffers the most severe negative mental health, physical health and academic outcomes (Kowalski & Limber, 2013). To mitigate these negative outcomes young people may be able to draw upon resources in their environment. Research has found a number of factors that protect those involved in online and offline harassment or bullying. Individual factors such as high academic performance and use of effective coping strategies are protective factors for depression associated with offline bullying (Hemphill, Tollit, & Herrenkohl, 2013). Studies have also found that social support is an effective buffer for the negative outcomes associated with offline bullying, specifically peer (Flaspohler, 2009; Holt & Espelage, 2007), family (Bowes, Maughan, Caspi, Moffitt, & 2 Arseneault, 2010; Conners ‐Burrow, Johnson, Whiteside ‐Mansell, McKelvey, & Gargus, 2009), and teacher or school support (Davidson & Demaray, 2007; Konishi, Hymel, Zumbo, & Zhen, 2010). The body of research examining buffers for online victimization is smaller than that for offline bullying and harassment. Research has found that strong parental relationships, positive school experiences (Davis & Koepke, 2015), and the ability to control one’s emotions (Hemphill & Heerde, 2014) have been found to decrease online victimization over time. To better understand how social support can be beneficial to those who are involved in online harassment further investigation is needed. Purpose of Study In this dissertation I examine the development of online behaviors over time. Young people involved in harassing others may learn their behaviors from peers or other models in their environments. It is important to understand what factors may influence the development of these behaviors as well as factors that may increase the likelihood of cessation of online harassment and victimization. Past research of offline harassment and bullying indicates that social support can buffer the negative effects of victimization, and therefore in this dissertation I examine the relationship among online behaviors and social support from teachers, peers, and teachers. In this chapter I provide definitions for terms that I use as well as outline the theoretical frameworks that inform this study. Definitions In this dissertation, I use a number of terms found in relevant literature, and to ensure clarity I provide definitions for use. 3 Harassers-Only Harassers-Only are people who harassed others but did not report being victimized. In the literature, it is not always clear whether a harasser was victimized as well because victimization may not have been measured in that particular study. In those cases, harassers are considered harassers-only for the purposes of reviewing the literature. Victims-Only Victims-only refers to those who have only experienced victimization and did not report harassing or bullying others. However, similar to harassers-only, it is not always clear whether victims harassed others as well as harassment may have not been measured. Harasser-Victims Harasser-victims are those who experienced victimization from and harassed others, specifically online. In the reviewed literature the most common terminology is cyberbully- victim; however, to stay consistent with terminology regarding online victimization and harassment, I use the term harasser-victims when discussing online interactions. Bully-Victims Because so little literature has examined online harasser-victims, I draw from the offline literature to review those who engage in both harassment and victimization as a group. I use the term bully-victim as is often used in the literature to describe those who are bullied and bully others offline or face-to-face. The use of this term also helps offset the offline literature from that of the online literature to make clear when I am addressing each group. Uninvolved 4 A fourth group of adolescents are sometimes discussed in the literature who are neither victim nor harasser, uninvolved youth. This group has no direct involvement with online harassment or bullying. Though studies do not always separate those involved in cyberbullying or online victimization from those who experience victimization and harassment, I use victims-only, harassers-only, and harasser-victims to delineate these groups, pulling from the current body of online victimization and cyberbullying literature. Closer Examination of Online Victimization and Cyberbullying In this dissertation I use the term online victimization rather than cyberbullying; however, the term cyberbullying is the most commonly used term in the media and in research to describe incidents of interpersonal aggression online (Guynn & Stobart, 2013; Pearce, 2013). Addtionally, other terms have been utilized to describe experiences of aggression online such as online harassment (Bossler, Holt, & May, 2012; R. Moore, Guntupalli, & Lee, 2010; Ybarra, Mitchell, Wolak, & Finkelhor, 2006) or online aggression (Dempsey, Sulkowski, Dempsey, & Storch, 2011; Law, Shapka, & Olson, 2010). Further, no clear definition of cyberbullying or online victimization exists in the literature (Aoyama, Saxon, & Fearon, 2011; Bauman et al., 2013; Hinduja & Patchin, 2008; Kowalski et al., 2014; Kowalski, Limber, & Agatston, 2008; Li, 2006; Smith et al., 2008; Tokunaga, 2010). This inconsistency is further complicated by a lack of explicit definitions across many studies (for a review see: Kowalski et al., 2014), and for the studies that do provide definitions, a number of problems arise. One issue found in some studies of cyberbullying is the use of the word “bully” without a description of what constitutes bullying in an online environment, either in an explicit definition or as an operational definition (i.e. the wording of instruments used in measuring incidents of 5 cyberbullying) (e.g. Spears, Taddeo, Daly, Stretton, & Karklins, 2015). By not defining bullying, these definitions do not clearly delineate the activities online that constitute cyberbullying. In these cases, young people in the study may have varying ideas of what constitutes bullying, giving rise to measurement difficulties. Other studies use only a single item to measure online victimization (Modecki, Barber, & Vernon, 2013) adding to measurement issues as it is not clear if the single question has strong measurement reliability. Another difficulty in defining cyberbullying is the ever-changing nature of technology and internet usage. For example, young people in the past used chat rooms (Li, 2007; Tynes, Reynolds, & Greenfield, 2004) and MySpace (Pfeil, Arjan, & Zaphiris, 2009) to interact, but now young people often use Facebook and Twitter (Duggan & Smith, 2013), and online games (DePaolis & Williford, 2014). Some definitions have used specific devices or types of online communication (e.g. chat rooms) (Kowalski & Limber, 2007). However, as technology grows and changes new websites and ways of socializing online may come into favor changing how people are able to interact with one another online. Any definition of cyberbullying should be created with that in mind. Because technology is dynamic, definitions of cyberbullying must balance the specificity needed to make the definition useful with the breadth needed to stay relevant even as technology changes. Other definitions of cyberbullying have focused on behaviors exhibited by aggressors online, relying on characteristics of offline bullying (e.g. Cappadocia, Craig, & Pepler, 2013; Hemphill, Tollit, Kotevski, & Heerde, 2014; Sticca, Ruggieri, Alsaker, & Perren, 2013). One example of this type of definition is used by Smith and colleagues (2008): “an aggressive act or behavior that is carried out using electronic means by a group or an individual repeatedly and over time against a victim who cannot easily defend him or herself.” This definition is quite 6 similar to that of face-to-face bullying which has three necessary features: 1) a repetitive negative act 2) perpetrated towards a less powerful peer 3) meant to cause harm (Olweus, 1993). Because online victimization has grown out of the face-to-face bullying literature, it is logical to adapt the traditional offline bullying definition for online interactions. However, there is disagreement about using these same three attributes to define cyberbullying, especially the idea of repeated acts (Spears et al., 2015). Research has also found that cyberbullying and offline victimization are not the same, and that cyberbullying is not just an extension of the peer victimization that occurs on the school grounds (Smith et al., 2008; Varjas, Henrich, & Meyers, 2009; Ybarra, Diener-West, & Leaf, 2007). These findings bring the validity of this definition into question, and further study in needed to understand whether this definition captures the construct of cyberbullying. Due to variations across studies, efforts have been made to provide a single, practical definition (Langos, 2012), but since the field has not come to a consensus regarding this matter, I choose not to use this term. I instead use the term online victimization to include cyberbullying and online harassment that may not constitute bullying. This more general and inclusive term is broad enough to encompass the many definitions that have been utilized under the term cyberbullying in the literature. Further, it is important to understand negative experiences that do not constitute bullying. Finally, the term online victimization centers on the victim rather than the aggressor as victims have less agency in the online interaction and therefore should be the focus of study. The definition I utilize for online victimization is: “threats or other offensive or rude behavior targeted directly to youth through technological channels (e.g., Internet, text messaging) or posted online about victims for others to see” (Finkelhor, Mitchell, & Wolak, 2000). 7 In this dissertation, I examine how online victim, harasser, harasser-victim, and uninvolved adolescents behave over time. Specifically whether young people shift between groups (e.g. uninvolved to harasser) and whether these shifts or transitions are related to social support from family, friends, or teachers. The study is informed by social cognitive theory and risk and resilience theory. Social cognitive theory describes the social process through which people learn behaviors such as online harassment of others. Risk and resilience theory provides a framework for understanding risk factors related to the later maladaptive or harmful outcomes and protective factors that may mitigate the effect of risk factors allowing people to have more positive outcomes in spite of risk. In the remainder of this chapter, I delineate these two theories in detail and situate the theories in peer victimization literature. Theoretical Framework Research has shown that peer victimization online is related to a number of negative social, psychological, and educational outcomes (Kowalski et al., 2014). There is also evidence that adolescents who are involved in online victimization as victim and/or harasser may not report engaging in the same behaviors later (Gradinger, Strohmeier, Schiller, Stefanek, & Spiel, 2012). Additionally, being an online harasser is a risk factor for later online victimization, indicating that neither victim nor harasser status is static. These changes may be due to a number of factors in one’s environment or within the self. social cognitive theory can be used to understand how adolescents develop online harassment behaviors and how the behaviors may change over time (Bandura, 1986). Further, factors that may mitigate the negative effects can be understood through risk and resilience (Garmezy, Masten, & Tellegen, 1984). 8 Social Cognitive Theory Social cognitive theory posits that people learn through the reciprocal interactions among personal factors, environmental factors, and behaviors (Bandura, 1986). The interactions among these three factors is known as triadic reciprocal causation and which informs a person’s learning (Bussey & Bandura, 1999). People can learn through feedback in the environment resulting from their own behaviors or by another person modeling a specific behavior and then analyzing the resulting consequences. Bandura posits that learning can occur from vicarious learning or, in other words, learning through observing the actions of others (Bandura, 1986). One observes the behaviors of other people in the environment and appraises the outcomes or consequences of those actions. Models for behaviors can be peers, adults, or even fictional characters. If the results of the observed behavior are beneficial toward attaining a goal, then the person may emulate the behavior in an effort to reach that goal as well. For example, if an adolescent witnesses a peer use aggression to improve his or her own social status, the adolescent who witnessed the aggression may follow suit in an attempt to better his or her own social status. However, observing a behavior is not enough. In order to emulate a modeled behavior, one must have the skills needed to perform the behavior and be motivated to behave that way as well. Social Cognitive Theory and Peer Victimization. Social cognitive theory has been applied in understanding aggression (Bandura, 1978) as well as offline peer victimization (Graham & Juvonen, 1998; Toblin, Schwartz, Hopmeyer Gorman, & Abou-ezzeddine, 2005). Children and adolescents learn to harass others by watching their peers harass others. Further, peers who model harassment receive beneficial outcomes such as bettered state through the 9 exercise of frustration leading them to continue their behaviors (Perry, Perry, & Rasmussen, 1986). Bullies also view aggressive behavior more favorably and as a more appropriate behavior than those who do not bully others (Burton, Florell, & Wygant, 2013). Further, bullies receive positive reinforcement from peers in the environment which leads to future bullying behaviors (Espelage, Holt, & Henkel, 2003). Studies examining online peer victimization have found similar results indicating that social cognitive theory may be useful in understanding how young people become online harassers. Evidence shows that online harassers are more likely than non-harassers to endorse aggression as a socially acceptable behavior (Williams & Guerra, 2007) and have fewer prosocial peer influencers (Cappadocia et al., 2013). Additionally, online harassers report more violent media exposure (Chang et al., 2014) and report feeling efficacious about their technological skills needed to engage in harassment (Walrave & Heirman, 2011). Additionally, since online interactions are often public (e.g. Twitter, Facebook, or in online gaming environments), adolescents have multiple opportunities to observe modeled harassment behaviors. This is evidenced by research which indicates that a greater number of adolescents report witnessing online harassment of others than actually experiencing harassment themselves (Patchin & Hinduja, 2006). Therefore, many adolescents have models for online aggression and harassment. Then, depending on the outcomes or reinforcement from others, they may continue or discontinue the behaviors. Social cognitive theory gives a framework through which the acquisition and cessation of online harassment of others may be understood. These behaviors may also develop or cease as young people get older because, as people age, they receive more feedback from the environment either reinforcing or diminishing particular behaviors through the reciprocal nature relationship 10 among the self, the environment, and one’s behaviors. Social cognitive theory explains the processes through which social factors influences learning. Research indicates that adolescents are often exposed to harassment of others online even if not victimized themselves; however, it is not clear which social forces may reinforce or inhibit online harassment behaviors. To better examine these relationships, I turn to risk and resilience theory. Risk and Resilience Theory Risk and resilience theory provides a theoretical grounding to understand the differential effects of online harassment and victimization. This framework has three core elements: risk, resilience, and promotive factors. Risk factors may be individual or environmental and add to a person’s likelihood for a particular negative outcome (Farrington, Loeber, & Ttofi, 2011). Resilience is the process by which a person overcomes risk factors and develops positively, and promotive factors are the individual or environmental characteristics that work against the effects of risk. The concepts of risk and resilience have been used to understand differential developmental outcomes. Early researchers of human development examined risk factors that led to maladaptive or negative outcomes for youth (Masten & Powell, 2003), often examining early characteristics of young people that acted as precursors for mental illness and psychological well-being (Garmezy, 1985; Masten & Garmezy, 1985; Rutter, 1979, 1985). In this section, I delineate the tenants of risk and resilience and outline the current research examining risk and resilience in research addressing online victimization. Risk. As the name suggests, one of the core components of the risk and resilience framework is risk. Early researchers found young people who experience more risk are more likely to experience adverse social (Christiansen & Evans, 2005; Wachs, 2012) and psychological outcomes (Mossakowski, 2003; Ragan & McGlashan, 1986), these may be a result 11 of a single risk factor or multiple sources of risk. Risk factors that are directly related to a single outcome are specific risk factors. For example, growing up with an alcoholic parent is a specific risk factor for later alcoholism, as those with alcoholic parents are more likely to be alcoholics than those without alcoholic parents (Ellis, Zucker, & Fitzgerald, 1997). Conversely, risk factors that are not related to a particular outcome are nonspecific. Growing up in a family with low socioeconomic status (SES) is nonspecific risk factor. Early poverty has been linked to a number of adverse outcomes, such as lack of access to medical care (Bradley & Corwyn, 2002), inadequate schooling (Evans, Marte, Betts, & Silliman, 2001), and mental disorders (Patel & Kleinman, 2003). Some of these outcomes could then lead to further negative life events. For example, poor access to medical care can lead to long-term medical issues in adulthood (Forsdahl, 1977), creating a cascading effect of risk, which serves to widen the effect of low SES on development and life outcomes. People with multiple risk factors may also experience the cumulative effect of risk. Research has shown that adolescents who are exposed to multiple risk factors early are more vulnerable in later life (Deater-Deckard, Dodge, Bates, & Pettit, 1998). Risk factors may be directly or indirectly related to negative outcomes and may collect over the course of a person’s lifetime, adding to the possibility of negative outcomes. Risk is complex and its factors varied. They can be specific or nonspecific and they can be long term or acute. Long-term risk may impact development though multiple developmental stages making this type of risk complex (Rosenthal, Feiring, & Taska, 2003; Williams, Davey, & Klock-Powell, 2003). Acute risk may be a single event; however, that does not mean that the impact is lessened (Bonanno, 2004). The sudden death of a parent is an acute risk that may have lasting psychological and social effects (Haine, Ayers, Sandler, Wolchik, & Weyer, 2003). Risk factors are important in understanding later negative outcomes for adolescents. 12 Risk factors of involvement in online victimization and/or harassment. In light of the negative consequences for all involved in online victimization as harasser, victim, or both, researchers have specifically examined risk factors for involvement in these online behaviors. Many studies have examined correlates of online victimization or harassment of others cross- sectionally; however, longitudinal designs are necessary to consider a correlate a risk factor as risk must precede the involvement in online victimization. These factors may be psychological, social or behavioral. Psychological. Though a number of studies have found psychological outcomes associated with online victimization and harassment, fewer studies have examined psychological risk factors for these behaviors. However, researchers who have studied this type of risk have found that depression (Cappadocia et al., 2013; Gámez-Guadix, Orue, Smith, & Calvete, 2013; Modecki et al., 2013; Rose & Tynes, in press; Yang et al., 2013), self-esteem (Modecki et al., 2013; Yang et al., 2013), anxiety (Yang et al., 2013), and exposure to media violence (Fanti, Demetriou, & Hawa, 2012) were risk factors for both online victimization and online harassment. Additionally, depressive symptoms were found to be a risk factor for later victim- harassment (Gámez-Guadix et al., 2013). Research also shows that callous-unemotional psychopathic traits are related to later online harassment (Fanti et al., 2012). Though only a few risk factors have been identified, there does seem to be a link between mental health and online victimization and harassment. Since mental health is a difficult issue to address systemically or through large scale interventions, it may more helpful to understand what online and offline behaviors may influence online victimization and harassment. Behavioral. Studies have shown that behavior online and offline may be related to later involvement in online victimization or harassment. For online harassment, researchers have 13 found a number of behavioral risk factors such as anti-social behaviors (Cappadocia et al., 2013), rule-breaking behaviors, greater amount of online communication (Sticca et al., 2013), and offline relational aggression (Hemphill et al., 2012). Research has also found that general offline problematic behavior (Modecki et al., 2013) higher amount of media exposure, especially anti- social media exposure, were related to later online victimization (den Hamer & Konijn, 2015). Further, Chang and colleagues (2014) found that more time playing online games, exposure to violence in the media, and risky internet behaviors the previous year was related to both online harassment and victimization, indicating that rather than only considering time online, types of internet use are important considerations in understanding risk. These studies found risk factors for victimization and for harassment; however, most did not consider whether adolescents engaged in both harassment and victimization at the same time as harasser-victims, though a small body of research has found some risk factors for this group. Research has found behavioral risk factors for the harasser-victim group (Cappadocia et al., 2013; Gámez-Guadix et al., 2013; Hemphill & Heerde, 2014). For this group offline bullying (Hemphill & Heerde, 2014) and offline victimization (Cappadocia et al., 2013) were found to be risk factors. Gámez-Guadix, Orue, Smith, & Calvete (2013) compared harasser-victims, victims, and uninvolved adolescents on a number of risk factors and found that harasser-victims reported higher levels of problematic internet use and substance abuse in the prior six months than the other groups. In the same study, victims were found to report more problematic internet use but no significantly different amount substance use at time one than uninvolved adolescents. By comparing groups on amount of risk, this study shows that in addition to differentially poor outcomes as a result of online victim-harassment, these young people may also experience 14 greater amounts of certain risk factors. Further research is needed to understand how these three groups, victims, harassers and harasser-victims, vary in amount of risk they may experience. Social. Social factors such as interpersonal relationships and social statuses have been found to be risk factors for some young people for either online victimization or harassment. Though peer support has been found to help mitigate the negative outcomes from peer victimization, socially preferred or popular students have been found to be at higher risk of involvement as victim and bully (Festl, Scharkow, & Quandt, 2014). Interestingly other studies have found that prior online victimization (Gradinger et al., 2012) and online harassment (Wegge, Vandebosch, Eggermont, & Pabian, 2014), especially for girls (Badaly, Kelly, Schwartz, & Dabney-Lieras, 2013), were related to later social popularity. Young people’s peer groups may also be risk factors for future victimization. Having fewer prosocial peers is a risk factor for later online victimization (Cappadocia et al., 2013). Further, classroom experiences seem to be risk factors as pro-cyberbullying attitudes of a student’s class were more predictive of perpetration of online harassment than individual pro-cyberbullying attitudes (Festl et al., 2014). These results show that influencers in one’s social environment may help shape the behaviors of young people. These findings seem to be in line with the social cognitive framework through which adolescents may develop online victimization behaviors. However, these findings do not show how supportive peers may influence the development of harassment behaviors or how peer support may be related to victimization. Further, it is unclear in these findings how family members and other adults may influence peer victimization. Further research is needed to understand how social support from adult sources may influence involvement in online peer harassment. 15 While research has shown that young people who face more adversity are at greater risk, not all young people exposed to the same risks have deleterious outcomes (Levin, Zigmond, & Birch, 1985; Morrison & Cosden, 1997; Tynes, Umaña-Taylor, Rose, Lin, & Anderson, 2012). This is due to protective factors or buffers in the environment that provide some shielding from the negative impacts of risk. Resilience. Resilience is seen as an interaction between people and their environment: “the capacity, processes, or outcomes of successful adaptation in the context of significant threats to function or development” (Masten, Best, & Garmezy, 1990, p. 426). There are three models of resiliency: compensatory, challenge, and protective (Fergus & Zimmerman, 2005). Compensatory models of resilience involve a direct effect of a promotive factor on an outcome (Zimmerman & Arunkumar, 1994). These factors work in direct opposition to risk on an outcome. An example of a compensatory factor would be a tutoring service in a low income neighborhood in which most students are at risk for lower educational attainment. For a compensatory model, as the amount of the promotive factor increases, risk decreases. The tutoring would act directly upon the educational outcome for young people with risk, poverty. In a challenge model very small or very high doses of risk lead to negative outcomes; however, with a moderate dose, a person does not experience the negative outcome. An example of this would be family conflict. If a person grows up in a house with no conflict, he or she may not learn to deal with conflict in later life. However, on the opposite end of the spectrum, high levels of family conflict may be abusive or create instability growing up which could lead to later familial abuse or instability. Instead, moderate levels of conflict would allow the child to learn conflict resolution which would be a positive outcome. The final is a protective factor model. In this model the protective factor moderates the relationship between risk and negative outcomes. 16 For example, a young person with a learning disability has a higher risk of dropping out of high school than a student with no learning disability (Levin et al., 1985); however, with higher levels of social and educational support at school (protective factors) these students have an increased likelihood of graduation (Morrison & Cosden, 1997). Further, there are two types of protective factors, protective-reactive and protective-stabilizing (Luthar, Cicchetti, & Becker, 2000). Protective-reactive factors act to lessen the negative outcomes associated with risk. Protective stabilizing-protective factors are moderating factors that neutralize the negative effect of risk on an outcome. In this study I examine protective-stabilizing factors in predicting changes in behaviors over time, such as cessation of harassment behaviors. Similar to risk factors, promotive factors in a person’s life may be multiple. These factors could be working together, or each could be independently working to reduce negative outcomes. For example, research has shown that young people who have support from peers and those with familial support have fewer negative outcomes from bullying (Bollmer, Milich, Harris, & Maras, 2005; Donnon, 2010; Fanti et al., 2012). It is unlikely that these promotive factors happen singularly though promotive factors are often examined individually in studies. Another consideration in examining promotive factors is multidimensionality in promotive factors. For example peer support has generally been found to be compensatory in the negative psychological outcomes associated with offline bullying (Davidson & Demaray, 2007; Holt & Espelage, 2007). However, when examining types of peer relationships, high quality friendships are more protective than low quality friendships (Bollmer et al., 2005). Promotive factors, like risk factors are complicated and may vary across a population. Finally, when examining promotive factors one may find that a factor is detrimental to outcomes. Instead of lessening the effect of risk as the three previous models illustrated, these 17 factors have an enhancing effect of negative outcomes. If young people who are at risk for depression were given a drug that increases the chances of suicidal ideation, the drug would have an enhancing effect for the relationship between depression and suicidal ideation. Though enhancing effects are not intended findings, understanding what factors may increase the effect of risk are important for building intervention programs to address risk factors in young people’s lives. Protective factors and resilience in peer victimization. Finally a risk and resilience framework can elucidate the resources in a young person’s life that can be used to combat the ill effects of peer victimization, both online and offline. Though a number of protective factors and coping mechanisms have been found, there is a significant body of work focusing on types of social support as protective factors for online and offline victimization especially in regards to familial and peer support. For example, supportive mothers who show warmth have been found to buffer the ill effects of peer victimization (Bowes et al., 2010), and adolescents with supportive parents who use few punitive measures are less likely to experience peer victimization offline (Baldry & Farrington, 2005). Despite the research exhibiting the protective nature of familial support Holt and Esplage (2007) found that peer support, but not maternal support were buffers for anxiety and depression for both victims and bully-victims. Research examining peer support has also been somewhat mixed. A few studies found that peer support was not related to a reduction in online victimization (Aoyama et al., 2011; Wang, Iannotti, & Nansel, 2009), while other research has found that it is related to a decrease in online victimization and harassment (Burton et al., 2013; Machmutow, Perren, Sticca, & Alsaker, 2012). Other researchers have found that peer support is a buffer for offline victimization (Bollmer et al., 2005; Hugh-Jones & Smith, 1999; Schmidt & Bagwell, 2007) and 18 having more socially adept peers in particular is a protective factor for negative outcomes associated with peer victimization (Tu, Erath, & Flanagan, 2012). Though the body of research of protective factors for later involvement in peer harassment is quite small, it seems that social support may be an important protective factor for online harassment and victimization. Though there is evidence that social support from one’s family is related to better psychosocial outcomes and less involvement in online harassment, the evidence regarding peer support is less clear. Further research is needed to address the inconsistencies in the literature. Further, these studies have examined victimization and harassment, but none have considered those who transitioned from victim to harasser-victim, for example. It is possible that considering those who are involved in later harassment-only or victimization-only as well as those who become involved in harassment-victimization, could elucidate differences in the peer support findings in the literature. Social support has been found to be an important environmental resource for adolescents involved in online harassment, but further examination is needed to understand how different groups of young people may react to higher levels of support from peers and adults in their environments. 19 CHAPTER 2: LITERATURE REVIEW Online victimization is problematic for all involved as victim and/or harasser. The consequences may vary by type of involvement and therefore protective factors found in young people’s environments may affect each group differently. First, to understand the consequences of each type of behavior, I outline the unique outcomes for those involved as victim, harasser, and harasser-victim. I then delineate issues in the literature in measurement arguing that the literature may be unclear due to methodological issues and assumptions. Consequences of Involvement in Online Victimization Experiences of online victimization are associated with negative psychological well- being through increased levels of depressive symptoms (Hemphill, Kotevski, & Heerde, 2015; Mitchell, Ybarra, & Finkelhor, 2007), more suicidal ideation (Bonanno & Hymel, 2013; Wang, Nansel, & Iannotti, 2011), lower self-esteem (Patchin & Hinduja, 2010), and anxiety in adolescents (Klomek, Marrocco, Kleinman, Schonfeld, & Gould, 2008; Mitchell et al., 2007; Tynes, Giang, Williams, & Thompson, 2008). There is also a relationship between emotional distress and victimization such as anger and sadness towards the offender, psychosomatic difficulties, and trouble sleeping (Patchin & Hinduja, 2006; Sourander et al., 2010; Ybarra & Mitchell, 2004). Finally, young people who report online victimization also report issues at school such as skipping class, detentions or suspensions, and even bringing guns to school (Hinduja & Patchin, 2007; Ybarra et al., 2007). These negative outcomes show that experiencing victimization online is quite harmful for victims overall. In the next section, I examine outcomes for online harassers who too experience negative outcomes. 20 Harassers-Only Young people who harass others online also report negative psychological and social problems. Researchers have found that online harassers have more depressive symptoms, more anxiety, and lower levels of self-esteem (Kowalski, Limber, & Agatston, 2012; Kowalski & Limber, 2013) than those not involved in victimization or harassment. They report less peer rejection than those who are victimized online, but more rejection than those uninvolved in online harassment or victimization. Online perpetration is also related to more offline aggression, (Steffgen, König, Pfetsch, & Melzer, 2011), more verbal aggression, and offline risky behaviors such as drinking or smoking cigarettes (Roberto, Eden, Savage, Ramos-Salazar, & Deiss, 2014). Online harassers also have trouble in school, reporting less school commitment (Ybarra & Mitchell, 2004), lower grades, and greater dislike for school (Kowalski & Limber, 2013). Many of the negative factors reported by victims are similar to those reported by their harassers, but victims report higher levels of these negative outcomes. Though harassers may not experience outcomes as detrimental as victims, these groups do seem to fare better than those who are involved in both harassment and victimization, harasser-victims. Harasser-Victims The harasser-victim group suffers the worst consequences in face-to-face peer victimization and online victimization (Fanti & Kimonis, 2013; Holt & Espelage, 2007; O'Brennan, Bradshaw, & Sawyer, 2009). Studies have found that the offline bully-victim group is more likely to endorse retaliation and show more aggressive-impulsive behaviors (O'Brennan et al., 2009), report higher levels of depressive symptoms, anxiety, and psychosomatic issues (Kaltiala-Heino, Rimpelä, Rantanen, & Rimpelä, 2000), more problem behaviors and poor school related functioning (Haynie et al., 2001). Similar to traditional face-to-face bully-victims, 21 online harasser-victims are at risk for negative outcomes (Sourander et al., 2010). An early examination of online harasser-victims by Ybarra and Mitchell (2004) found that this group is unique within the online victimization dynamic. Specifically, online harasser-victims displayed more depressive symptomatology and engaged in more problem behaviors than youth who were only victims or harassers. Online harasser-victims also reported more anxiety, suicidal ideation, depressive symptoms, lower self-esteem and grades (Kowalski & Limber, 2007, 2013). Though adolescents who are online harasser-victims suffer the worst psychosocial outcomes, their outcomes are not completely unrelated to the victims-only and harassers-only as all three groups suffer poor psychological well-being and social exclusion though harasser-victims report more severe experiences of these shared negative outcomes (Sourander et al., 2010). Overall, these studies show that teens involved in online victimization suffer negative consequences, but those who are online harasser-victims tend to have more psychological and social issues than any other group. Further, research has shown that the online harasser-victim group may be more common than the victim or harasser only groups, making this group particularly important to understand through further research (Mishna, Khoury-Kassabri, Gadalla, & Daciuk, 2012). Researchers have examined whether onset of being a harasser-victim happens when both victimization and harassment of others happens concurrently or whether one behavior precedes the other. Some research shows that in offline interactions, the victimization comes before the harassment (Ma, 2001). This could mean that some who are victimized harass others as retaliation. A study by Werner, Bumpus, and Rock (2010) found that adolescents who reported online victimization were 16 times more likely to be online aggressors themselves compared to those who reported no online victimization. Another study by Jose, Kljakovic, Scheib, and Notter (2012) found that for adolescents harassing others online was related to later victimization 22 online. However, this study did not examine online harasser-victims specifically. It is unclear whether those who harassed became victims-only or harasser-victims. Further research is needed to understand the relationship between harassment and later victimization online. Transitions among Harasser and Victim Groups Over Time Studies of online victimization often examine victimization, harassment, or victim- harassment separately; however, there is evidence that people may engage in different online behaviors over time. Some longitudinal research has found that those who are online harassers or victims often continue the same online behavior over time as online victim, harasser, or harasser- victim (Gámez-Guadix et al., 2013; Hemphill & Heerde, 2014; Sticca et al., 2013); however, these studies do not indicate that all participants involved at time one are still involved in the same behaviors at later time points. Further, Gradinger and colleagues (2012) found that online harassment was not stable over the time of one year as online harassment was not predicted by harassment the year prior. Though results are somewhat muddled, this research shows that some adolescents change their behavior patterns regarding involvement in online harassment over time. Research has shown that earlier online harassment may lead to later online victimization for some young people (Wright & Li, 2012). However, there is less evidence supporting a transition for victim to harasser, with some studies finding the relationship to not be significant (Hemphill & Heerde, 2014; Yang et al., 2013). A study by Chang and colleagues (2014) examined how Taiwanese high school students transitions between either harasser or victim and uninvolved status over a year. They found that adolescents who were either victim or harasser in tenth grade were more likely to be uninvolved by eleventh grade. The authors attributed these shifts to developmental changes, but not all young people in the study followed the same 23 behavior patterns as 10.6% who were uninvolved in tenth grade reported online victimization in eleventh grade. Though many adolescents do not change their behavior, there are some who do. Because of the negative effects of involvement in victimization in any role, it is important to understand what factors may be more likely to lead adolescents to become uninvolved and understand what factors may be related to becoming involved, or specifically becoming a harasser-victim as this group has the most deleterious outcomes. Understanding factors that may lead to more positive outcomes can inform interventions aimed at minimizing adolescent involvement in online victimization. Overall, the research examining group differences has found varying outcomes by gender and age. The measurement of the cyberbullying construct varies across studies. This lack of clarity in group differences may be due to methodological issues in the online victimization literature, and in the next section I delineate these methodological concerns. Methodological Issues The prevalence rate of online victimization varies across studies (Kowalski et al., 2014; Tokunaga, 2010). This may be due to differences in the sample or it may be due to inconsistent measurement. Without a clear definition, researchers may be measuring a wide array of constructs under the term cyberbullying. Also, research has shown that the methods used to ask participants about online victimization can influence the ways participants answer the questions (Ybarra, Boyd, Korchmaros, & Oppenheim, 2012). In this section, I further examine these issues within this body of literature. 24 Prevalence rates of victimization Many studies have found prevalence rates of online victimization between 10–40% across samples of adolescents (Hinduja & Patchin, 2008; Li, 2006; Patchin & Hinduja, 2006; Smith et al., 2008; Ybarra & Mitchell, 2008). Though most studies fall in this range, there are some studies that have found rates well outside of this range (Beran & Li, 2007; Juvonen & Gross, 2008; Mishna, Cook, Gadalla, Daciuk, & Solomon, 2010). For example, Juvonen and Gross (2008) found that 78% of participants in their study had been cyberbullied online. Participants were asked about “mean things” experienced online instead of using the term bullying or describing specific behaviors they deemed cyberbullying. It is likely that the broad definition of cyberbullying used in their measure lead to the discrepant findings. Additionally, studies that have found particularly low prevalence rates are often national studies that employed random digit dialers to recruit their sample (Jones, Mitchell, & Finkelhor, 2012; Wolak, Mitchell, & Finkelhor, 2007). In these studies, random homes in the US were called and an adolescent was surveyed over the phone. Since parents were contacted first to give consent, they were aware of the content of the interview. If their parents were in the room or nearby, the adolescents in the study may have responded less truthfully to the questions about sensitive topics—such as their own behaviors online—for fear of consequences. Further, the method of data collection may have influenced the rates as research has shown that people are less likely to be forthcoming about sensitive topics when surveyed over the phone or in person rather than on a computer (Evans, Kessler, Lewis, Peters, & Sharp, 2004). Another factor that could impact the prevalence rates of victimization is timeframe of the of the victimization measure. The measures utilized by researchers often give a particular time 25 constraint for the victimization they are measuring. Participants have been asked about victimization in the last three months (Wade & Beran, 2011), the last six months (Jones et al., 2012), or last year (Mitchell, Finkelhor, Wolak, Ybarra, & Turner, 2011; Tynes, Rose, & Williams, 2010; Ybarra et al., 2007). Some studies have even asked about the participants’ lifetime experiences or gave no timeframe at all (Hinduja & Patchin, 2007; Varjas et al., 2009). These varying ranges coupled with inconsistencies in definitions likely drive the lack of agreement across studies measuring the prevalence of online victimization or cyberbullying. Measurement The lack of a succinct definition for cyberbullying has also lead to inconsistencies in the types of questions used across measures. Some studies use survey instruments with multiple questions assessing a participant’s experiences online (Mitchell et al., 2007; Rose & Tynes, 2015), others simply ask participants if they have been cyberbullied with a single question (Li, 2006; Moore, Huebner, & Hills, 2012; Raskauskas, Gregory, Harvey, Rifshana, & Evans, 2010). For example, in a study by Li (2006), participants were asked to circle yes or no in response to the following statement on a survey as a measure of student experience with online bullying: “I have been cyberbullied (e.g. via email, chat room, cell phone)” (p. 170). All follow up questions in the survey asked participants to select from a list the ways in which cyberbullying had taken place. By not operationalizing the term cyberbullying for the participants in the study, the researcher likely measured a different construct for each participant. This is evidenced by Menesini and colleagues’ (2012) study that examined whether people across European countries operationalized the term cyberbullying similarly by asking people whether they would label examples of online behavior as cyberbullying. They found significant variation across countries, showing that people have different constructions of the term’s meaning. 26 Though giving participants no common definition is problematic, even using the term cyberbullying in the question changes the way participants may respond to surveys (Ybarra et al., 2012). Ybarra and colleagues found that, participants were more likely to label described online behavior as cyberbullying when neither the word cyberbullying nor a definition of cyberbullying was provided to participants. Methods of measurement, timeframes of measured victimization, as well as language used in measuring cyberbullying and online victimization vary across studies. As the field moves forward, these measurement issues should be addressed to clear up the inconsistencies across studies. A final issue in the online victimization literature is the lack of longitudinal studies. By studying online victimization over time, instead of cross-sectionally, researchers are better able to examine the factors related to online victimization. Specifically, longitudinal designs could move the field forward in attempting to understand what factors are outcomes from victimization and what factors may be risks for online victimization for young people. For example, an individual who is a social outcast may be a target for online bullying, but online bullying could also make an individual a social outcast. By measuring individuals across time points, researchers will more clearly understand the temporal order of events. Also, examining the same participants over time would allow researchers to see how victimization affects young people through adolesence which will help clarify scholarly discussions regarding the effect of age on online victimization. Additionally, these studies could examine the ways social support and other positive factors in a young person’s life may help alleviate some of the negative effects of online victimization over time. 27 CHAPTER 3: METHOD Since past research shows that online harassment is related to detrimental outcomes for youth, it is important to examine these negative effects and possible protective factors for young people. Current research examining online harassment has often examined victims, harassers, and harasser-victims based on the offline literature; however, there is little research that has investigated whether these groups best describe online victimization and harassment as well. To address these gaps, the current study explores the underlying online groups in the victim/harasser dynamic and to test protective factors, particularly social support from adults and peers, of racially diverse adolescents. The current longitudinal study will provide better a better understanding of the ways social support may influence later online behaviors giving insight into ways families, friends, and teachers may be able to decrease the likelihood of further involvement in online victimization and/or harassment. The research questions that drive this study are: 1. Are there distinct groups of adolescent online victimization and harassment behavior? Do adolescents transition between groups over time? 2. Does social support from teachers, peers (online and offline), or family predict initial patterns and changes in online victimization participation over time? 3. Does social support from teachers, peers (online and offline), or family predict transitions between statuses over time? At the outset of the study I formulated hypotheses for these research questions. For the first question, I expected to find groups in the data that align with the past online and offline bullying literature. In other words I expect to find harasser-only, victim-only, harasser-victim, and uninvolved groups. I also hypothesized that a majority of participants will remain in the same 28 group between each wave. However, I also expected that some participants will transition to other groups as has been found in the literature (Chang et al., 2013; Gámez-Guadix et al., 2013; Hemphill & Heerde, 2014; Sticca et al., 2013). For my second research question, I hypothesized that offline peer, teacher, and family support will be related to a higher likelihood of being less involved in online victimization and/or harassment at wave one as these factors have been found to be related to less victimization (Baldry & Farrington, 2005; Davidson & Demaray, 2007; Konishi et al., 2010; Machmutow et al., 2012).. However, I expected online peer support to be related to more victimization at time one. This is because of research indicating that spending more time online is a risk factor for victimization and those who seek out support online are likely spending more time on the internet than their peers. Finally, for question three, I expected each of the social support variables to be protective. Past research has found that offline peer support (Burton et al., 2013), and familial support (Bowes et al., 2010) to be protective for the ill effects of online victimization, and teacher and school support have been found to be protective for offline victimization (Davidson & Demaray, 2007; Konishi et al., 2010). Though past research has not examined the nature of online peer support as a protective factor for online victimization and harassment, research has shown that some marginalized youth, such as lesbian, gay or bisexual (LGB) youth look online for support that they do not find in their offline environment (Hillier, Mitchell, & Ybarra, 2012). Therefore it is possible that those who seek out more online friendships do so to find emotional support which could be protective. Data Project Description For this study, I used an existing dataset collected over a three-year period through the Teen Life Online and in Schools Project. The project aimed to understand online victimization 29 and possible buffers for young people focusing specifically on experiences of racial discrimination online and offline. Because the project was particularly concerned with issues of racial discrimination, recruitment of participants was designed to create a racially diverse sample with a majority of adolescents of color. Students completed surveys that included a number of psychological adjustment variables such as depression, anxiety, and empathy. The protective factors included measures of racial and ethnic identity and social support from peers and adults. Racial discrimination offline and online were measured, and more specifically the online discrimination measure also included online sexual harassment and general online harassment. Because the survey was comprised of a wide array of measures assessed longitudinally this dataset is useful in studying general online victimization as well. This dataset was collected during the fall semesters of 2010, 2011, 2012, and 2013 from K-8, middle and high school students from twelve schools—both public and charter—in the U.S. Midwestern region. Most schools in the dataset were somewhat racially diverse with eight schools having no more than 60% of one racial/ethnic group. Of the remaining four schools, three had over 98% black or African American students enrolled and one had more than 80% Latino students enrolled. Though the data collection period spanned four years participants were only measured three times. For all participants there is one year between each wave of data collection. The initial year included 236 participants from two schools. The next year added 789 more participants and ten more schools. The small group of participants measured in year one were measured for wave two in year two when the 789 were measured for their wave one, and then measured for wave three in year three when the other participants were measured for wave two. Procedure. Research assistants recruited students from classrooms that were selected by administrators based on access to laptops or computers labs. The classes selected were primarily 30 technology, English, and homeroom. Parental consent forms and fliers were distributed to approximately 150 students per school with copies available in English and Spanish. Adolescents in middle school grades and early high school grades were targeted to increase the odds that the participants would still attend the school during the three years of the study. The research team did, however, maintain contact with graduated students through email correspondence. During distribution, research assistants gave a brief 10-minute presentation to the selected classes to describe the purpose of the study. On a prearranged date, research assistants returned to each school to administer surveys via web link to all students who returned the affirmative parental consent forms. Online surveys were sent to email addresses that were provided by participants, which were accessed during the allotted classroom time. In the event that participants did not have a valid email address prior to survey administration, temporary email addresses were established for survey access. In a small number of cases, surveys were accessed via a web link. Once access was granted, survey administration occurred over one to two consecutive class periods to give participants ample time to finish the survey at school. Research assistants were present to inform students of confidentiality, explain terms, and troubleshoot any technical difficulties. Prior to beginning the survey, the research team explained that the respondents would be asked about their online experiences and their feelings about themselves. They were informed of confidentiality and told that they had the right to stop at any point in the survey. Following the completion or termination of the survey, all students were provided with resources such as local counseling services and internet safety websites to report online predators. As an incentive, students received $15 Amazon.com gift certificates for their participation for wave one, $20 in wave two, and $25 for wave three. Further, the research team 31 maintained contact during the study through birthday emails to remind students about the project and encourage participation in each of the follow-up years. Additionally, participating schools were provided with a small stipend. Recruitment and consent procedures were reviewed by the Institutional Review Board of the principal investigator’s institution. Participants. The data included a racially diverse group of young people from the Midwestern U.S. Wave one included 1028 adolescents (46.4% male, 53.6% female) in grades 6- 12, with ages ranging from 10 – 18 (M = 14.47, SD = 1.88). Wave two included 743 participants and wave three 559 participants. Based on youth self-report, the ethnic-racial distribution of the sample was 32.8% African American (n =337), 25.6% European American (n = 263), 22.7% Latino (n = 233), 7.0% multiracial, and 11.9% Other or unidentified (n=123). Participants also reported the highest attained education levels of their parents. Participants reported that 34.9% of their fathers received an undergraduate or graduate degree (n =359), 25.6% a high school diploma (n =263), 4.2% did not finish high school (n =44), and 35% were missing or reported that they did not know (n =360). Highest maternal education levels were 41.7% undergraduate or graduate degree (n =429), 25.3% high school diploma (n =260), 4.3% did not finish high school (n =44), and 28.7% were missing or reported that they did not know (n =295). Parental consent was required for participation, and the average response rate among the twelve schools was 49.8%. This figure is based on the total number of consent forms returned from those distributed to students, not on active refusals. Victimization and Harassment Variables Online Victimization. This is measured by the general harassment subscale of the Online Victimization Scale (OVS)(Tynes et al., 2010). This subscale measure has four items measuring types of victimization online. In the current study this full four-item scale had a 32 reliability coefficient of α=0.84. Two items were used for the online victimization items in analyses: “People have posted mean or rude things about me on the Internet” and “I have been harassed or bothered online because of something that happened at school”. Items were scored on a scale from 0= “Never” to 4= “Everyday”. Harassing Others Online. Online harassment was measured by three questions. Two questions measured whether or not participants engaged in harassment during the previous year. The third question asked participants to endorse items on a five-item checklist listing reasons for which they harassed someone online (e.g. “Because someone called you a name”). The first two items were scaled 1= “Never” to 6= “Everyday”. The checklist only asked participants to endorse items that were true for themselves, so these items are coded as 1=”Endorsed” and as missing otherwise. The victimization and harassment scales have a total of 12 items, so the number used needed to be pared down in order to create a more parsimonious model and to limit the number of parameters estimated in the model especially as covariates were added to the model (Collins & Lanza, 2010). To choose the items for analyses, I first selected the items on the OVS scale that encompassed different dimensions of victimization. The victimization items: “People have posted mean or rude things about me on the Internet” and “I have been harassed or bothered online because of something that happened at school” address incidents that begin both online and offline. The harassment items from the checklist were chosen because they parallel the victimization scale. Three of the listed items from the harassment checklist were used in the analyses: “Because someone called you a name,” “Because something that happened at school,” and “Because of the way the person acts”. These three items were selected because of the five checklist items they had the highest rates of endorsement. 33 [Insert Table 3.1 about here] After selecting the victimization and harassment items, I dichotomized them for clear interpretation of the analyses. For the victimization variables, items were recoded zero for an answer of “Never” on that item and one for all other response numbers as they indicate experiences of harassment at some point in the previous year. In the dataset, the harassment variables were only coded as one if the participant endorsed the item and missing if the participant did not chose to endorse or skipped the item. Therefore, I needed to determine whether participants had chosen not to endorse an item, or had skipped the question. To identify the reason for a missing response, either not endorsed or skipped, I examined the responses to the items in the complete harassment measure. The harassment checklist included four reasons for harassing others (ex. “Because someone called you a name”) and one final item indicating no involvement: “Didn’t call someone a name”. Based on participants’ non-response on these items they were classified as either missing or not endorsed. Any participant who did not answer the first two questions in the scale (“How often in the past year have you harassed or bothered someone online?” “How often in the past year have you called someone a bad or nasty name online?”), and did not endorse any of the items on the checklist in question three were marked as missing (W1=6.3%, W2=1.2%, W3=1.4%). If a participant endorsed any item or items on the checklist, for the remaining items on the checklist all missing responses were coded as not endorsed. Next, any participants who responded in the first two questions indicating that they had not participated in harassment online and did not endorse any items on the checklist were coded as 0=not endorsed as they indicated no involvement in online harassment and endorsing any item would be antithetical to these responses (W1=7.1%, W2=7.8%, W3=8.6%). Finally, participants who did not endorse any 34 items on the checklist but indicated that they had engaged in harassment on the first two questions were marked as missing because not endorsing any reason for their behaviors indicated that they likely skipped the checklist in the question (W1=6.3%, W2=1.4%, W3=1.4%). Social Support Variables Social Support from Peers, Teachers, and Family. These variables were measured using a revised version of a combination of two scales: the Multidimensional Scale of Perceived Social Support (MSPSS) (Zimet, Dahlem, Zimet, & Farley, 1988) and the Child and Adolescent Social Support Scale (CASSS) (Malecki & Demaray, 2002). The combined scale includes 12 items with three four-item subscales to assess social support from peers, teachers, and family. The MSPSS contains three subscales each measuring a different type of social support: family, friends, or significant other. The significant other questions were taken out and replaced with the questions from the CASSS that measure teacher support. Example items from each scale include: “My family really tries to help me”, “I can count on my friends when things go wrong”, and “My teacher(s) is fair to me.” Responses range from 1=”Strongly Disagree” to 7=”Strongly Agree”. The reliability coefficients for the peer, family and teacher supports for the current sample are α=0.93, α=0.90, and α=0.87 respectively. Online Peer Support. This variable was measured by a single item. Participants were asked how often they sought support online: “When I have a problem, I discuss it with friends online”. Responses ranged from 1=”Never” to 5=”Very Often”. Missing Data Missing data is particularly problematic in Latent Transition Analysis (LTA) modeling. In Mplus 6.11, the program I used for the LTA analyses, it is possible to use full information maximum likelihood (FIML) to estimate LTA models with no covariates (Muthén & Muthén, 1998-2010). However, when covariates are added to the model any missingness on those 35 covariates is dealt with only by listwise deletion instead of FIML. This use of listwise deletion would bias the estimates (Schafer & Graham, 2002). Because of this bias I decided to use multiple imputations in order to complete the model estimations without missing data. The original dataset had missing data through both attrition and incomplete surveys. The reasons for dropout may be due to measured or unmeasured factors. The dataset has a number of variables; therefore, it may be possible to find characteristics of those who dropped out to create a model of missingness from which the data can be imputed to reduce bias due to attrition. The attrition rate for the study across waves was somewhat high. Between waves one and two 28% of participants did not return, and between waves two and three 25% of participants did not return. In total, from wave one to wave three 46% of participants dropped out of the study. To better understand the patterns of missingness I performed Decision Tree Analysis using the rpart package in R version 3.1.1 to find the variables in the dataset that best explain attrition for the study. DTA is particularly useful in exploring data as it allows for non-linearity and interactions among predictors (McArdle, 2013). The results from these analyses were used to create a model of missingness and determine the variables to be used for multiple imputations. To explore the data attrition, I first created a variable of missingness. The study uses three waves of data, so I created a variable in which 1=missing by wave three and 0=participated in all waves. I then entered all demographic variables—age, gender, school attended at wave one, and race; all variables used in this study—the full OVS scale, online harassment scale and social support variables; peer victimization in-school; and mean values from available psychological scales in the dataset including depression, self-esteem, anxiety, and academic motivation. These variables were then used to create a decision tree predicting missingness by wave three. To make sure the model did not over fit the data, I pruned the tree back by selecting the number of splits 36 associated with the lowest cross validation (Zhang, 1999), see Figure 3.1. Pruning the model back balances the complexity of the model with model fit to the data. The model with three splits was selected and the final model is shown in Figure 3.2. [Insert Figure 3.1 about here] [Insert Figure 3.2 about here] DTA indicated that attendance at two particular schools, one high school and one middle school, and age were related to missingness. The middle school participated in the first year of data collection but was not willing to allow our research team to administer surveys in follow-up years. However, since surveys links were sent to participants via email, we reached out to all participants in the school during follow-up years to attempt to collect data. Though some participants did return, this school had higher attrition than all other schools. This school was 82% Latino, 7.7% white, and 7.2% African American. The high school participated in data collection for three waves and was more racially diverse with 30.2% Latino, 28.6% white and 32.9% African American student body. Although the middle school had a majority of Latino students, being Latino was not related to missingness. The third variable related to missingness was age, with younger participants more likely to drop out than older participants. This was somewhat surprising because some older participants graduated during data collection. The research team reached out in following years for participation through email, and it seems these participants were not more likely to drop out of the study than younger participants who did not graduate. After variables related to missingness were determined, social support variables were standardized for ease of interpretation of odds ratios in later analyses. Using Stata 12 (StataCorp, 2011) the dataset was imputed with multiple imputation using chained equations. The model 37 included school of attendance, age, and all variables used in the LTA model. Twenty datasets were imputed for analyses as this has been found to be an adequate number of datasets for a modest amount of missing data (Graham, Olchowski, & Gilreath, 2007). Analyses LTA was selected for this study for a number of reasons. Like all latent modeling, this type of analysis accounts for error in measurement (Collins & Lanza, 2010). This is important to consider since measured constructs may have measurement error which cannot be directly observed or calculated. Further, LTA uses a person-centered approach. While variable-centered analyses examine the relationships among variables and expand those findings to the population, a person centered approach examines changes in the individual as a whole (Bergman, Magnusson, & El Khouri, 2003). By examining the data through person-centered analyses, the latent groups are defined by the patterns in the data rather than using predetermined groups. Studies of online harassment and victimization have often used the groupings found in the offline literature; however, it is possible that these are not accurate descriptions of harassment online as these are different behaviors and experiences. Examining the underlying groups in the data precludes making these assumptions. Further, the online literature has often examined bullying online, which is a specific type of harassment. It is possible that the common groups found in bullying are not accurate when examining the more inclusive construct of harassment. Another strength of LTA modeling is the flexibility to add covariates to the model in order to examine predictors of initial classes and change in classes over time (Collins & Lanza, 2010). To test significance of covariates, the model with the covariate is compared to the measurement model without the covariate using a likelihood ratio χ 2 test. LTA modeling is useful in examining the predictive nature of variables such as social support in the initial patterns 38 of behaviors young people engage in as well as the association between social support and transitions between statuses over time. LTA also has a few weaknesses. Because the groups or statuses are determined by examining the data, the method is exploratory in nature. However, I have elected to use LTA because so little work has examined the unique groups in online victimization and harassment especially longitudinally. Further, the analyses examining the relationship between the covariates and transitions may have issues as well. Because the number of participants who make a particular transition may be small, regression coefficients for the relationship between the covariate and transitions may have low power and therefore unreliable coefficients. To combat this, I examined the number of participants at each transition and decided if the number of participants was too small for interpretation. In spite of these weaknesses, LTA can provide a nuanced examination of the changes in online harassment and/or victimization behavior of adolescents over time. 39 CHAPTER 4: RESULTS Descriptive Statistics I explored descriptive statistics of the data using the dichotomized harassment and victimization variables and the unstandardized social support variables. Means and standard deviations of the variables indicated low endorsement of both victimization and harassment items as the means for these items fell between 0.1 and 0.4 (see Table 4.1). This is consistent with findings in the literature indicating that about a third of online users experience some type of online harassment (Kowalski et al., 2014). Examining each variable over time, endorsement of victimization variables increased over time but all other variables remained mostly stable over the three waves. The means for the social support variables indicate overall high levels of offline social support reported by participants. Though the online support variable is measured on a smaller scale—1 to 5 instead of 1 to 7—it is still clear that participants indicated less online social support overall. This may be because the day-to-day interactions with family, teachers, and friends may provide some of the support indicated by these findings whereas adolescents may need to reach out to find online support. Also, not all participants may spend time online communicating with peers, but face-to-face interactions naturally occur within each family or school. [Insert Table 4.1 about here] A correlation matrix showed that the victimization and harassment variables were positively correlated with one another (Table 4.2). The social support variables were negatively correlated with the victimization and harassment variables except for the online peer support variable which was positively correlated with the victimization and harassment items. Therefore online peer support may be related to harassment and victimization differently than the other 40 support variables. Examining the magnitudes of the correlations shows that the support variables are not strongly correlated with any one victimization or harassment variable; however offline friend support has a particularly small correlation with the harassment and victimization variables. It seems that those who experience victimization or harassment may have generally low levels of online support. Further analyses will examine how these types of support are related to harassment and victimization. [Insert Table 4.2 about here] Model Building In LTA, item response probabilities are determined for each status. Since all items were dichotomized, probabilities can be interpreted in terms of endorsed or not endorsed. Interpretation of these probabilities was based on model fit and theory that was derived from the literature. To examine how many classes accurately describe the data, I estimated an LCA model for each wave with an increasing number of classes, beginning with two and stopping when a model with one added class is no longer a better fit to the data then the previous model. Because the models are not statistically nested, the log likelihood ratio difference test is not a useful test for differences in model fit (Collins & Lanza, 2010). Instead, the Bayesian Information Criterion (BIC; Schwarz, 1978) and the Akaike Information Criterion (AIC; Akaike, 1987) are two reliable statistics when considering the number of classes in a latent class model (Collins & Lanza, 2010). For both the AIC and BIC, values closer to zero indicate a better fit, so as fit improves, the BIC and AIC decrease. Models were fit to the data with increasing classes until the AIC and BIC increased. In the case that the two disagreed, the most theoretically sound model was chosen. [Insert Table 4.3 about here] 41 The fit statistics for each of the LCA models is listed in Table 4.3. At wave one, the AIC and BIC did not agree on the preferred model. The AIC determined that a three class model was sufficient, and the BIC indicates that 5 classes are needed. Because these statistics do not agree, I examined the item response probabilities in each class across the three, four and five class models to determine which made the greatest amount of theoretical sense. [Insert Figure 4.1 about here] Figure 4.1 illustrates the item response probabilities for each class. For these probabilities, a high number indicates higher endorsement of an item and a lower number, lower endorsement. In the three class model there seems to be a victim, harasser, and uninvolved class. The four class model seemed to contain a victim, harasser, harasser-victim, and uninvolved group. However, in class one—the harasser-victim class—probabilities were not as expected. The harassment items and the victim1 item all indicated high probabilities, but the victim 2 item did not. Because the class is high on the harassment items and one of the victimization items, I still interpret this as a harasser-victim class. Looking at the five class model, I found that class two were victims, class five were harasser-victims, class three were harassers, and class four were uninvolved, but the fifth class was less easily interpretable. In this final class the harasser 1 item had a probability close to .5 while the other items indicated uninvolvement in the other harass and victim items. This item is “because someone called you a name” which could make this a retaliatory class, but since this class is low in victimization, it is hard to establish what they are retaliating against. After examining each of the three models I decided that the four class model had the clearest class interpretations, and it follows the theoretical assumptions that similar to offline 42 harassment there are victims-only, harassers-only, harasser-victims, and uninvolved youth. Further, it is aligned with the LCA models for waves two and three data. After determining the number classes using LCA, I constructed an LTA model to fit to the data. Because the three waves had previously been analyzed separately, I assessed model fit for the full LTA model. Three waves of data were used and an initial model in which item probabilities were allowed to vary across waves was fit to the data. This model was compared to a time invariant model in which each item was constrained to have equal item response probabilities at each wave. The χ 2 distribution of the loglikelihood difference was examined to determine if the constrained model was a significantly different fit to the data, which would indicate that the model is not invariant over time. Results indicated that the model fits were significantly different (χ 2 (diff) =82.58 df=38 p>0.001). Time invariance is important for interpretation of transitions over time as this indicates that the statuses cannot be interpreted the same way at each wave, and since my research questions require time invariance I tested a two wave model to see if invariance would hold. Because the third wave of data had so many imputed data points I elected to drop the third wave and tested a two wave model with wave one and wave two data. Results indicated no significant difference between the constrained and unconstrained models with two waves (χ 2 (diff) =29.25 df=21 p>0.108). Therefore, all further LTA analyses use the two wave model with item response probabilities constrained to be equal at both waves. 43 Question 1: Are there distinct groups of adolescent online victimization and harassment behavior? Do adolescents transition between groups over time? The LCA models used in the model building indicated that distinct groups of harassment and victimization do exist in the data. At each wave the four classes indicated a victim-only, harasser-only, harasser-victim, and uninvolved classes—see Table 4.4. The parallel nature of the groups at each of the three waves indicated that these classes are useful classifications of online victimization and/or harassment. [Insert Table 4.4 about here] After fitting the two wave LTA model to the data, I examined the item response probabilities in the model to ensure that the statuses were still victim-only, harasser-only, harasser-victim, and uninvolved. The first status is high on all items and is labeled the harasser- victim status. Though the victim 2 item has a lower probability than the other items, it is high enough to show that participants in this status are likely to endorse the item. Status two is a clear victim-only class. Though the harassment items in status three do have some variability, comparing this status to the other three statuses, it appears that the victimization items are more similar to the last status with overall low endorsement while the harassment items are higher endorsed. The general pattern of the status is still a higher probability of harassment than victimization, so this is still interpreted as a harasser-only status. The fourth and final class is an uninvolved class with all probabilities quite low. This indicates that there is a harasser-only, victim-only, harasser-victim, and uninvolved status in the data. Once the latent statuses were determined, I examined the number of participants in each class. Because the dataset is 20 imputed datasets, all estimates are averaged over all the datasets, and the participant counts for each status are not round numbers. At wave one more than half of 44 participants, 57.9%, were in the uninvolved status, and at wave two 47.9% were uninvolved. The second most common status at waves one and two is harasser-only with 28.0% and 14.1% at each respective wave. The harasser-victim and victim-only statuses increased in proportion between waves one and two. The harasser-victim status had 8.8% at wave one and 14.1% at wave two, and the victim-only status increased from 5.3% at wave one to 23.9% at wave two. Interestingly, these two groups both included victimization as a component, however; further analysis was needed to understand the patterns of change for these two statuses over the two waves. I next examined the patterns of change through the transition probabilities from the LTA model without covariates (see Table 4.5). The diagonals in the table represent the probability of staying in the same status between waves. For victims and uninvolved statuses, there seems to be little movement out of either status between wave one and wave two. Victims-only had an 86% probability of staying a victim and uninvolved participants at wave one had a 74% probability of staying uninvolved at wave two. For the harassers and harasser-victims, there was 53.9% and 40.2% probability, respectively, of staying in the same status at wave two. [Insert Table 4.5 about here] Next I examined the latent transition probabilities for patterns of how participants changed statuses over the two waves. The patterns in the probabilities of change showed that participants were more likely to cease harassment behaviors than to escape victimization by wave two. Harasser-victims, for example, had a 34% probability terminating harassment behaviors and becoming victims-only by wave two. Harassers-only who did change status were likely to become victims as participants in this status had a 21% probability of becoming victims-only and 27% probability of becoming harasser-victims at wave two. For the 45 participants who experienced victimization, the probabilities of moving to groups without victimization were quite low. The harasser-victim status only had a 3% probability of becoming a harasser-only and a 9% probability of becoming uninvolved, and the victim-only status had a probability of .2% of moving to the harasser-only status—which is very close to zero—and a probability of 5% of becoming uninvolved. For harassers-only, there seemed to be more movement between statues at one and two, though they were still most likely to stay harassers-only. For this group, the probability of becoming a harasser-victim was similar to the probability of becoming uninvolved 21% and 16% respectively indicating that though many young people experienced more victimization over time, some were still able to move to an uninvolved status and out of the victim/harasser dynamic. However, the harassers-only at wave one were more likely to be involved as victim- only or harasser-only at wave two than to become either harasser-victim or uninvolved. The harasser-victim group also showed some movement—less than the harassers-only— and a higher probability of staying in a victimization status than becoming harasser-only or uninvolved. This again shows that harassment behavior seems to be more likely to cease, but victimization was likely to continue over time. Interestingly the transition with the second greatest probability for harassers-only was to victim-only. Finally, the uninvolved group was highly likely to stay uninvolved and seem to have a higher probability of becoming victims (15%) than becoming either harasser-victims (5%) or harassers-only (6%), though only slightly. Examination of the transition probabilities supports my hypotheses and shows that participants do change statuses over time. However, for each status, the probability of staying in the same status at wave two had a higher probability than transitioning to a different status (e.g. harassers-only was more likely to stay harassers-only than transition to another status at wave 46 two). Interestingly the second most likely transition was to victim-only. For those who reported victimization-only at wave one, the two most probable statuses at wave two involve victimization. It seems that harassment may be more temporary than victimization. These findings show the particular challenge for young people to escape harassment online, and the possibility of stopping harassment behaviors. The rest the analyses aim to explore factors related to the changes in behaviors of participants to determine possible interventions for online harassment and victimization. Question 2: Does social support from teachers, peers (online and offline), or family predict initial patterns in online victimization participation? To examine the relationship between the social support variables and initial status, I added each of the social support variables to the LTA model as predictors of wave one latent status membership only. Before entering any covariates each variable was standardized for ease of interpretation. To assess the significance of the covariate, I used a likelihood-ratio χ 2 test to compare the model without covariates to each model with a covariate of interest. This tested the null hypothesis that the vector of three logistic regression coefficients related to the covariate in the model—β 1 —is equal to zero. Therefore a significant χ 2 difference indicated that the covariate was a significant predictor in the model. Results showed that each covariate was significant, see Table 4.6. [Insert Table 4.6 about here] For each analysis a logistic regression coefficient was estimated for each status except for a reference status. I used the uninvolved status as the reference in each model since ideally, young people would not be involved in online harassment. This allows comparison between the likelihood of an unwanted status—any of the three types of involvement—to the preferred status 47 of uninvolved. Table 4.6 gives the logistic regression coefficients and odds ratios. Odds ratios are the exponentiated logistic regression coefficient or e β 1 which can be interpreted as the odds of membership in a particular status compared to the uninvolved status when the covariate is increased by one standard deviation above the mean (Collins & Lanza, 2010). Therefore, odds ratios greater than one indicate that a one standard deviation increase in social support is related to increased odds of membership in a particular status compared to the uninvolved status, and odds less than one indicate that a one standard deviation increase in social support is related to lower odds of membership in a status compared to the uninvolved status. Though I report both logits and odds ratios in the table, odds ratios can be more intuitive to interpret, so those are the values I used to describe the findings in models with covariates. Overall the results indicated that offline support from peers, teachers, and family members was associated with greater odds of membership in the uninvolved status, whereas higher amounts of online peer support was associated with an increased likelihood of harasser- victim, victim-only, or harasser-only status than uninvolved at wave one. Participants who reported higher levels of offline peer support were less likely to be either harassers-only (β= - 0.392 OR= 0.676), harasser-victims, (β= -0.198 OR=0.820), or victims-only (β= -0.606 OR= 0.546) at wave one than to be in the uninvolved status. However, online peer support is associated with greater odds of being harassers-only (β= 0.803 OR= 2.232), victims-only (β=0.312 OR= 1.366), or harasser-victims (β= 0.665 OR= 1.944) than uninvolved at wave one. This finding is cross-sectional so it is not possible to determine from these results whether online support leads to these behaviors or if the likelihood of membership in these three wave one statuses is correlational. 48 Teacher support was related to lower odds of membership in the victim-only (β= -1.071 OR= 0.343), harasser-only (β= -0.669 OR= 0.512), and harasser-victim (β= -0.489 OR= 0.613) class and family support was also related to decreased odds of participants being in the victim- only (β= -0.472 OR= 0.624), harasser-only (β= -0.895 OR= 0.409), and harasser-victims (β= - 0.576 OR= 0.562) statuses at wave one. Support from families, teachers, and peers offline were related a higher probability of membership in the uninvolved status at wave one, whereas support from peers online was not. However, these findings are only cross-sectional because the levels of support were measured concurrently with the victimization and harassment items. To examine the relationship between social support and later online harassment and/or victimization, covariates were added to the model to predict wave two status conditional on wave one status. These findings are addressed in the next research question. Question 3: Does social support from teachers, peers (online and offline), or family predict transitions between statuses over time? To examine this research question, I added each covariate to the model to predict the transitions from wave one to wave two instead of wave one status only. Each covariate was entered separately to assess the individual relationship between that variable and the change of status. These results are presented as odds ratios for ease of interpretation; however in these models, the reference status was staying in the same status between waves one and two. Additionally, because the probability of transition between victim and harasser was very close to zero (0.02), coefficients for the covariates were not estimated for this transition and the transition probability was constrained to zero in each of the models. Results for the models showed that all social support covariates were significant predictors. Because the model with the covariate added a constraint to the transition between victim-only and harasser-only, the covariate model was 49 compared the baseline LTA model with the same constraint. The loglikelihood ratio χ 2 difference and odds ratios are reported in Table 4.7. Finally, in interpreting these results it is important to note that some of the cells of the transition tables include a small number of participants, see Table 4.7. The harasser-victim to harasser-only cell contains an average of 3.4 participants and the harasser-victim to uninvolved transition only contains an average of 8.3 participants across the 20 imputed datasets. The victim-only group to the harasser-victims, harassers-only, and uninvolved cells contain 7.1, 0.1, and 3.0 participants on average. As noted earlier, the victim- only to harasser-only transition is not estimated in any model and because the transition probability is very close to zero. Coefficients were estimated for the other cells with less than ten participants; however, the coefficients and odds ratios may not be reliable. I reported all coefficients in the results tables but did not interpret findings the small cells to err on the side of caution. The results table notes the small cells. [Insert Table 4.7 about here] Online peer support model was significant (χ 2 (diff) = 59.864, df=11, p < 0.001). Results show that harasser-victims with a one standard deviation increase in online peer support were less likely to transition from the harasser-victim status to the victim-only status at wave two (β= - 0.092 OR= 0.912). Those in the harasser-only status with higher online peer support were more likely to transition to the victim-only (β= 0.231 OR= 1.260) or the harasser-victim (β= 0.545 OR= 1.725) and, interestingly, the uninvolved status (β= 0.375 OR= 1.455) at wave two. Increased online peer support for the uninvolved participants at wave one were more likely to transition to harasser-only (β= 0.233 OR= 1.250), victim-only (β= 0.125 OR= 1.133), and the harasser-victim (β= 0.150 OR= 1.162) statuses than to stay uninvolved at wave two. These findings indicate that harasser-victims with higher online peer support are likely to remain 50 harasser-victims than transition to only victimization indicating that this group did not stop harassment easily. Harassers-only were more likely to transition to the harasser-victim status which is associated with more detrimental outcomes in the online victimization literature. However, harassers-only are also more likely to transition out of that status to the uninvolved status with increased online peer support. For those in the uninvolved status at wave one, higher levels of online peer support are related to higher odds of transitioning to more negative statuses, harasser-victim, victims-only and harassers-only. [Insert Table 4.8 about here] The offline peer support model was also significant (χ 2 (diff) = 46.966, df=11, p < 0.001). The results indicated that for a one standard deviation increase in peer support harasser-victims at wave one are equally likely to remain harasser-victims than to transition to victims only (β= 0.001 OR= 1.001). An increase in peer support is also related to an increased likelihood of harassers-only transitioning to the harasser-victim (β= 0.361 OR= 1.435), victim (β= 0.118 OR= 1.125), or uninvolved status (β= 0.399 OR= 1.490). For those in the uninvolved status at wave one with an increase in peer support have about the same odds of staying uninvolved at wave two as transitioning to the victim-only status (β= 0.028 OR= 1.028) and lower odds of transitioning to the harasser-victim status (β= -0.438 OR= 0.645) or harasser-only status (β= - 0.644 OR=0.525) at wave two. The teacher support model had a significant loglikelihood ratio χ 2 difference test (χ 2 (diff) = 96.344, df=11, p < 0.001) indicating that teacher support is a significant predictor for the transitions between waves. A one standard deviation increase in teacher support is related to lower odds of transitioning out of the harasser-victim status at wave one to the victim-only status (β= -0.631 OR= 0.532). Harassers were found to be more likely to transition to the harasser- 51 victim (β= 0.203 OR= 1.225) and the victim-only (β= 0.114 OR= 1.121) statuses for a one standard deviation increase in teacher support at wave one. Additionally, a one standard deviation increase in teacher support is related to slightly lower odds of transitioning to the uninvolved status (β= -0.023 OR= 0.977). Finally, the uninvolved participants were less likely to transition to harassers-only (β= -0.498 OR= 0.608) and slightly less likely to transition to the victim-only status (β= -.052 OR= 0.949) than to stay uninvolved with a one standard deviation increase in teacher support. With a one standard deviation increase in teacher support the odds of transitioning to the harasser-victim status from the uninvolved status (β= 0.356 OR= 1.428) were higher than staying in the uninvolved status at wave two. The final model examined family support as a predictor of transitions between waves one and two. I first fit the model with the transition between victim-only to harasser-only constrained and no coefficient for that transition. The model was not identified. The beta parameter for the regression of family support on the transition from uninvolved to harasser-only was indicated as the problem in estimation. Further examination of the error indicated that the problem was small variance in family support for participants in that particular transition. I adjusted the model and removed the regression of family support on the uninvolved to harasser-only transition. This model fit the data and the loglikelihood χ 2 difference test indicated that the family support variable was significant (χ 2 (diff) = 75.392, df=10, p < 0.001). Results showed that the odds of staying in the harasser-victim status are about equal to the odds of transitioning to the victim-only status (β= 0.004 OR= 1.004) at wave two for a one standard deviation increase in family support. Harassers-only at wave one with a one standard deviation increase in family support are more likely to transition to the harasser-victim (β= 0.388 OR= 1.474), victim-only (β= 0.127 OR= 1.135) and uninvolved statuses (β= 0.419 OR= 1.520). 52 For uninvolved participants at wave one, the odds of staying uninvolved are about equal to the odds of becoming a victim-only (β= 0.038 OR= 1.039), and the odds of transitioning to the harasser-victim status are lower (β= -0.549 OR= 0.578) than staying uninvolved at wave two. The findings for this model show that higher levels of family support only increase the odds of transitioning out of harasser-only status. Further, harassers-only have higher odds of transitioning to other negative online behaviors and ending their involvement in online harassment. 53 CHAPTER 5: DISCUSSION The results from these analyses indicate that online victimization and harassment are dynamic issues for adolescents. Findings indicated that young people transitioned among victim and harassment groups over the course of a year. Initial statuses were related to social support and some changes were more likely for those with higher levels of social support. The findings of the relationship between the types of support and transitions are not straightforward and indicate a complicated relationship between online victimization and/or harassment and social support. In this chapter, I discuss the results and integrate these findings with theory to better describe the social influences of online harassment and victimization. I also delineate the limitations of this study and discuss the contributions this study makes to the literature. I then suggest future directions for further research and outline implications of this study for schools and educators. Summary of Findings My first research question addressed the underlying groups in online victimization and harassment as well as the changes in these groups over time. Examination of the cross-sectional LCA models indicated that there was likely a harasser-only, victim-only, harasser-victim, and uninvolved classes, even though the first wave of data was not as clear on the classes in the data. Putting the three waves of data together in an LTA model proved difficult. The item-response probabilities were not consistent over time. Dropping the third wave of data allowed the item- response probabilities to remain constant over the two waves and therefore the two wave model was utilized for analyses. The latent classes in the LTA model matched that of the LCA models and next the transitions among classes were examined to understand patterns of change. 54 Transition probabilities of the model without covariates supported my hypothesis that participants in each wave one status were most likely to stay in the same status at wave two rather than transition to a different status. This is a positive finding for the uninvolved group as there was a 73% chance of remaining in the uninvolved status. The harasser-only participants had a 36% probability of staying in the same status at wave two and had the highest probability of transitioning into the uninvolved status at 16%. The similarity among transition probabilities made harasser-only the most dynamic status. Conversely, victims-only were the most likely to remain in the same status at wave two, about an 81% probability, indicating that victimization is not a temporary issue but a sustained problem for some young people. The probability of moving from victim-only to harasser-only was very close to zero indicating that those who are able to escape online victimization are not likely to become harassers. The most likely transition out of the victim-only status was to the harasser-victim status. This finding may indicate that some victims attempt to retaliate against their harassers, but this retaliation does not seem to end their victimization, as there was nearly a zero percent probability of becoming a harasser-only. If retaliation did end harassment it would be logical that victims-only would become harassers-only in some cases. Finally, for victims-only, it seems that some were able to transition to the uninvolved status at wave two, but the mechanisms of that change are not clear as the number of participants who made that transition was too small to examine in analyses with covariates predicting transitions. Analyses also indicated that ending victimization was more difficult than ending harassment over the course of a year. Though harasser-victims tended to stay in the same status, the most common status change was to victim-only. This change means that some participants stopped harassment behaviors but were not able evade victimization. The harasser-victim group 55 had very low probabilities of becoming harassers-only or uninvolved indicating that they were likely to experience victimization the following year, even if they were no longer engaging in harassment online. Harassers-only were the least likely to stay in the same status at wave two and were more likely to become victims-only than harasser-victims. This indicates that the most likely status changes for those who harass others are to statuses in which participants experience victimization, and those who transition to victim statuses are more likely to stop harassing (i.e. victim-only) than continue harassing others (i.e. harasser-victim). Because harasser-victims have the most deleterious outcomes (Kowalski & Limber, 2007, 2013; Ybarra & Mitchell, 2004), researchers have examined the path through which young people become harasser-victims, but findings from past research have been mixed. Prior studies of offline victimization found that victimization often precedes harassment behaviors (Ma, 2001) but research has also found that online harassment precedes online victimization (Jose et al., 2012). The findings from this study indicate that both victims-only and harassers-only transition to the harasser-victim status at wave two. For victims-only, the most probable transition out of that status is to the harasser-victim status. Very few victims-only transitioned to the uninvolved status and the probability of transitioning to harasser-only was ostensibly zero. For harassers- only, the probability of transitioning to the harasser-victim status was similar to the probability of transitioning to the victim-only status. Therefore, patterns show that both harassers-only and victims-only are at risk for becoming harasser-victims indicating that there is no one pathway to becoming a harasser-victim. After examining the transitions in statuses over time, I examined the relationship between each wave one status and the social support variables. First, I entered each support variable individually in the LTA model to predict initial statuses in order to test whether social support 56 was related to online behaviors. Findings indicated that each of the social support variables was related to initial statuses. Offline peer support, family support, and teacher support variables related to greater odds of being uninvolved than any other status. However, online peer support was related to greater likelihood of harasser-only, victim-only, and harasser-victim statuses. This finding supports my hypothesis that, support from teachers, family, and offline peers is related to greater odds of being uninvolved in online victimization and/or harassment, and online peer support is related to lower odds of uninvolvement at wave one. Because these findings are cross- sectional it is not possible to discern causation. From these findings, it therefore seems that seeking out support online may be related to more risky online behaviors. This is consistent with prior research which found that online communication (Sticca et al., 2013) and increased exposure to media (den Hamer & Konijn, 2015) are risk factors for online victimization. After finding that social support was related to latent class membership at wave one, I next examined the relationship between the social support variables and each of the possible transitions. These findings were less clear than the prediction of initial group status. Harasser- only was the only wave one status that had enough participants to interpret findings for a transition to the uninvolved status at wave two. Analyses indicated that a one standard deviation increase in peer support online, peer support off line, and family support was each related to greater odds of transitioning from harasser to uninvolved which indicates that these types of social support may be protective for some adolescents. However, analyses also indicated that a one standard deviation increase in social support from peers online and offline, teachers, and family members is associated with greater odds of transitioning to a status with victimization— victim or harasser-victim—than staying in the harasser-only status. These findings show that the types of social support tested are not unidimensional; there seems to be variance in the influence 57 of social support across participants in the harasser-only status. Further, these contradictory findings have been found across other studies as well. Some studies have found that offline peer support was not related to a reduction in online victimization (Aoyama et al., 2011; Wang et al., 2009), and others have found that offline peer support is related to a decrease in online victimization and harassment (Burton et al., 2013; Machmutow et al., 2012). However, only the Machmutow and collegues (2012) study was longitudinal. Though the findings seem to conflict, the results of the current study may indicate that differences in the type of support adolescents are receiving from their peers is an important consideration. Previous studies have found that adolescents with fewer prosocial peers are at greater risk for online victimization (Cappadocia et al., 2013), which could account for the likelihood of transitioning to a status with victimization rather than staying in the harasser-only status for some participants. Other researchers have found that having more socially adept peers in particular is a protective factor for negative outcomes associated with offline peer victimization (Tu et al., 2012) Therefore, there may be a difference in the quality or type of support that leads to the discordant outcomes, and these findings show that supportive relationships are not enough. Turning to the uninvolved status at wave one, it seems that higher levels of offline peer and family support predict lower odds of transitioning to harasser-victim status than staying uninvolved and about equal odds of becoming a victim-only as staying uninvolved. These results indicate that these types of support are related to more favorable transitions for those who were not involved; however, for online peer and teacher support this was not the case. Higher levels of these two types of support were predictive of greater odds of becoming a harasser-victim than staying uninvolved. For a one standard deviation increase in online peer support, the odds of transitioning to the victim-only and harasser-only statuses were greater than the odds of staying 58 uninvolved. These findings indicate that these two types of support are risk factors for some adolescents. Even though higher online peer support was found to be predictive of greater odds of transitioning to the uninvolved status from the harasser-only status, this type of support is also risky for those who have avoided online harassment and/or victimization. Interestingly, higher levels of teacher support were associated with lower odds of transitioning to the harasser-only status than staying uninvolved. This indicates that teacher support may increase the risk of victimization for students, but deter students from harassing others. Past research of teacher support and online victimization has found that teachers are often not aware of the victimization specific students are experiencing (Cassidy, Brown, & Jackson, 2012). It is possible that teachers provide students with social support generally but do not provide support for online victimization specifically. Further, teacher support may be interpreted by peers as a weakness or indicate that a student is an easy target for victimization, increasing the likelihood of transitioning to a victimization status at higher levels of teacher support. Overall, these findings indicate that social support from peers, teachers, and family members are important considerations. Further study is needed to understand the characteristics of a supportive relationship that is protective rather than a risk factor. The findings also indicate that online victim-only, harasser-only, harasser-victim, and uninvolved statuses are not always stable. This is especially true for those who harass others as this behavior seems to be the most likely to change over the course of a year. In the next section, I apply my theoretical framework to the findings to further elucidate the results. Theoretical Implications of Findings This study was guided by two theories, social cognitive theory and risk and resilience. Social cognitive theory describes the social process through which people learn behaviors such 59 as online harassment of others. Risk and resilience theory provides a framework for understanding risk factors related to the later maladaptive or harmful outcomes and protective factors that may mitigate the effect of risk factors allowing people to have more positive outcomes in spite of risk. Social cognitive theory Social cognitive theory can be used to examine the learned behaviors of adolescents in the study. This theory has been used to explain the development of aggressive harassment behaviors through vicarious learning (Bandura, 1978; Graham & Juvonen, 1998). Examining the transition probabilities in the model without covariates indicates that those who are victimized may be learning and emulating these behaviors. For those who were victims-only at wave one, the most likely transition out of being a victim-only was to the harasser-victim status at wave two indicating that some of the victims-only retaliated or harassed others, though these efforts did not stop their victimization. However, it seems that most victims-only at wave one did not emulate harassment behaviors and remained victims-only at wave two. Additionally, the probability of those uninvolved at wave one staying uninvolved was quite high, but for those who did transition it seems that they were likely to experience victimization-only rather than to take up harassing others as harassers-only or harasser-victims. These findings indicate that harassment behaviors may not be widely emulated. Social reinforcement in some manner possibly through peers engaged in online harassment themselves may be needed, and to examine this relationship, covariates were added to the model to predict transitions between waves. Results from the models with covariates predicting transitions over time showed that higher offline peer support was related to a lower probability of uninvolved adolescents at wave one transitioning to either the harasser-only or harasser-victim status than staying uninvolved. 60 However, higher online peer support is related to greater odds of transitioning out of the uninvolved status to harasser-only, harasser-victim, and victim-only statuses. The differences in these findings may be due to the environment in which these social relationships are formed and maintained. Also, those who seek out online peer support communicate with these peers online which is a risk factor for online victimization (Sticca et al., 2013). Online communication may also expose adolescents to more online harassment and victimization either directly or indirectly, and this exposure could lead to vicarious learning of harassment behaviors online. Further adolescents who have been exposed to online harassment that receive reinforcement in these behaviors from peers may be more likely to develop harassment behaviors than those who do not have peers reinforcing these online behaviors. However, this distinction is unclear from the current study. Risk and resilience For adolescents in this study the highest probability for each status at wave one was to remain in the same status at wave two. Therefore, harassment, victimization and harassment- victimization are risk factors for membership in the same status at wave two. Those in the uninvolved status at wave one were not experiencing the risk factors measured in this study because they were not experiencing victimization or harassment, and therefore this group could not demonstrate resilience in this study. All four types of social support were used to predict transitions in order to examine adolescent resilience. Results showed support variables were both protective factors and risk factors in the study. Online peer support was a risk factor for uninvolved participants to become victims-only, harassers-only, and harasser-victims. In understanding this finding, it may be important to consider why adolescents reach out for online support. Young people who feel disenfranchised 61 or marginalized, such as lesbian, gay, or bisexual youth, seek out support and friendships online that they cannot find in their offline environment (Hillier et al., 2012). Further, research shows that in computer-mediated communication, adolescents are more likely to disclose intimate information, which could in turn be used to harass that adolescent later (Valkenburg, Schouten, & Peter, 2005). Therefore those who seek out online peer support may be at risk in other ways and may be engaging in more risky online behaviors. Further research is needed to examine the online experiences and individual characteristics of these young people. Though online peer support was a risk factor for some adolescents in the study, other types of social support were found to be a protective-stabilizing factor for others. In this study, a transition from a status of involvement in online victimization and/or harassment to the uninvolved status suggests resilience as these participants were in wave one statuses that were at risk for further harassment and/or victimization but transitioned to the uninvolved status. Because so few harasser-victims and victims-only transitioned to the uninvolved status at wave two, coefficients for social support on those transitions are not reliable and were not interpreted. However, the transition from harasser-only to uninvolved was considered and findings showed that online peer support, offline peer support, and family support were protective-stabilizing for some harasser-only participants as higher levels of these types of support were predictive of greater odds of those in the harasser-only status at wave one transitioning to the uninvolved status at wave two. This finding supports my hypothesis that social support would be protective for adolescents in the study. However, teacher support did not follow this pattern. Instead, this covariate had an enhancing effect. Participants who reported higher levels of support from their teachers had higher odds of staying in the harasser-only status than transitioning to the uninvolved status at wave two. This finding is problematic and contrary to my hypothesis, but 62 may be an important consideration as schools address the issues of online victimization and harassment. These findings indicate that actors in adolescents’ lives online and offline may be important resources in their lives. Though some of the findings indicate that social support can be a risk factor for involvement in online harassment and/or victimization, there is also evidence that support may be able to help young people end their involvement in online harassment. The differences in outcomes may be due to the type or quality of support young people receive. With further exploration of the differences in social support as a protective factor and a risk factor could help shape effective interventions for online harassment and/or victimization. Limitations This study, like all research, has some limitations due to the nature of the model and dataset. The results from this study are somewhat contradictory and indicate that further research is needed. This lack of clarity may be due to the nature of the analyses or limitations of the dataset. In this section I discuss these limitations to outline the limits of the results. One limitation of the study is the self-report data from an online survey. It is always possible in survey research that participants did not truthfully disclose information. However, research has shown that people are more likely to be forthcoming about sensitive topics such as past incidents of online victimization or harassment when surveyed on a computer rather than over the phone or in person (Evans et al., 2004). Further, self-report requires participants to evaluate their own experiences which can lead to inconsistencies across participants’ responses. To combat this problem the study used a questionnaire of online harassment rather than asking if participants had been victimized or cyberbullied; however, one participant may evaluate an experience differently than other participants. Further, for online victimization, asking parents, 63 peers, or teachers to report whether a participant is victimized or harasses others online could be problematic as well. Unlike classroom behaviors, it is unlikely that teachers and parents have access to the entirety of an adolescent’s online interactions. Though self-report data may capture online victimization and harassment fully, other methods of measurement have significant drawbacks. A second limitation of this study is the presence of missing data. In spite of the best efforts of the research team, some participants elected to not return to the study. In one case this was partially due to school administrators who did not want to participate in the study after the first year of data collection. Though efforts were made to recruit the students from this school to participate in follow up years, there was a significant loss of participants from that school. Another data issue concerns the online harassment items. The items were measured using a checklist and it was not clear for all participants whether they skipped the question or chose not to endorse each item on the scale. Responses to other questions were used to determine whether the checklist was skipped, but there is possibility of error. Finally, since the statistical program used for analyses was not able to use FIML to estimate all cases in the dataset once covariates were added to the model, multiple imputation was employed to handle the missing data. A model of missingness was created using decision tree analysis and variables from the survey as a whole were included in the model. Analyses indicated three variables most related to missingness: attendance at two particular schools and age. These variables were added to the model for multiple imputation of the dataset to reduce bias. However, it is possible that the missing data was related to variables not measured in the study. For example participants with low reading levels or short attention spans may have been less inclined to participate and complete the survey in follow up years as the survey was quite long. If there were unmeasured variables related to 64 missingness, the absence of these variables in imputing the dataset may lead to bias in the findings. Another limitation of this study is the exploratory nature of the design. Though the research questions and modeling was based on relevant theory, the examination of the underlying latent statuses is exploratory in nature. The statuses found at each of the three time points were aligned with the classes found in offline victimization literature, and in that way the results do confirm these past findings. However, the model was created by examining the data rather than using the data to test prescribed theory. Further, the scale did not hold time invariance for all waves and the third wave was dropped from the LTA model. Another limitation of the LTA model concerns significance of the covariates in the model. Significance testing of the covariates is done by examining the model fit with and without the covariate. If the model with the covariate has a significantly different loglikelihood ratio χ 2 difference test, the covariate is significant in the model, however, there is no test of significance for the odds ratios in the results. Because of this, the findings should be further tested in other datasets to examine whether the victim-only, harasser-only, harasser-victim, and uninvolved classes are present and distinct. Another limitation arose with the loss of the third time point. Because the third time point was dropped the timeframe for which changes in online harassment and/or victimization statuses could be examined. A third time point of data could have provided some informative analyses. For example, results of the current study show that some participants in the online victimization- only and harassment-only statuses changed to the harasser-victim status at wave two. With a third time wave of data, I could have examined whether these two groups were equally likely to stay harasser-victims at wave three or if one group was more likely to remain in the harasser- victim group adding a more complete understanding around the pathways adolescents take to the 65 harasser-victimization status online. Additionally, a third wave of data may have given some of the needed nuance in understanding social support as a predictor of transition probabilities over time. The ability to compare the results between two transition points could have shown consistencies and inconsistencies in findings as well as longer term outcomes between those who experienced social support as a risk factor and those who experience social support as a protective-stabilizing factor. When examining the changes in status over time it was clear that some transitions were more common than others. Therefore, some transitions were reported by only a few participants. The transitions with fewer than ten participants could not be analyzed in this study, and this precluded analyses of some of the transitions of interest. For example, from the current study findings it is not clear whether social support was related to the transition to uninvolvement for harasser-victims or victims-only. The transition probabilities in the model without covariates indicate that transitioning out of a victimization status was unlikely and it is unclear if social support was predictive of that desired change. This issue limits the information that can be gleaned about the protective nature of social support for the least common transitions. To address this, future analyses should use a larger dataset. This would increase the number of participants in each transition thereby increasing the number of participants for the less common transitions to a number large enough to reliably estimate in the multinomial logistic regression. Though the findings of this study indicate that social support is an important mechanism in understanding online harassment and/or support, this study is not causal and therefore it cannot be ascertained whether or not each type of social support caused the status transitions found in the study. It is possible that individual characteristics lead young people to receive more beneficial support from teachers or parents. Additionally, friendships are self-selected not 66 random and individual characteristics of adolescents may influence the types of peers from whom they seek support. Therefore the supportive or risky nature of social support may be related to other risk factors for online victimization and/or harassment. Because social support may vary in quality another limitation of this study is the lack of information about the qualities of the social support young people received. Offline peer support, teacher support, and family support were measured using larger scales and can pick up on multiple dimensions of social support; however, the online peer support measure is a single question. It is possible that the variation in findings for this variable may be in part due to the unidimensional manner in which the variable was measured. The findings around this variable should be considered in future work with a more nuanced examination of online peer support. Further examination of types and quality of support from parental relationships (e.g. nurturing vs authoritarian), peer relationships (e.g. prosocial vs antisocial peers), and teacher relationships (e.g. whether teacher knows about online victimization or harassment) could disentangle the seemingly conflicting results indicating that social support is both a risk and protective- stabilizing factor. Future Directions for Research Findings from this study indicate that supportive relationships are not adequate in preventing online harassment and/or victimization among adolescents. The findings from this study indicate that higher levels of social support from offline peers, teachers and family members is related to greater odds of uninvolvement cross-sectionally. However, the results examining the influence of social support over time show that for some participants higher social support was a risk factor while for others it was protective. Future research is needed to examine differences in social support across these groups. Further analyses should examine the intensity 67 and duration of the victimization or harassment in relation to social support. In this study, the recoding of the variables to either endorsed or not endorsed allowed clear interpretation of the latent statuses in the model, but precluded analyses of intensity and duration of negative online experiences. Future work should examine the relationship between level of support and intensity of online harassment, victimization and harassment-victimization. For some young people, finding friends in school or face-to-face is difficult and they may reach out to peers online. Studies suggest that some adolescents are able to get support they do not receive offline, while other work suggests that communicating online is a risk factor for victimization. If those who do not find peer support offline are able to find some support online, they may choose to continue these relationships in spite of negative online experiences. Further research is needed to understand the benefits of online relationships with respect to the possible victimization. Contributions to the Literature This study is the first of its kind to examine the changes in online victimization and/or harassment over time and the relationship of these changes to social support. First, this study provides evidence that there are distinct harasser-only, victim-only, harasser-victim, and uninvolved groups of youth online. Further, this study shows that even though membership in each group tends to be stable, there is change in how adolescents interact online, specifically that harassment is less stable than victimization over time. Additionally, the study is concerned with general online victimization and harassment and bullying rather than the more specific behaviors of cyberbullying and cybervictimization. This study also provides evidence that there are underlying groups within online victimization and harassment dynamics and that young people are able to move between groups over time. Finally, this study examined the relationship among 68 multiple types of support and victim and/or harassment statuses. The findings from this study provide new insights into the development and cessation of online harassment and victimization, and the importance of social support for adolescents who may be involved in these negative experiences. Overall, this study adds to the current body of online victimization and harassment literature in a number of important ways. While this study indicates that distinct groups exist and the data and that these groups are similar to those found in the bullying literature, one important contribution of this study is the examination of online victimization and harassment rather than just bullying and bullying victimization. Though these constructs overlap, harassment is a more global concept that may be experienced by more adolescents over time. Though the harmful effects of bullying make it an important research area, it is also important to understand how adolescents engage in and experience harassment as it is a more common, but still quite detrimental experience for youth. Within the experiences of online harassment and victimization I found four distinct groups. Because the participants were classified into one of the four groups, it was possible to examine victims, harassers, and harasser-victims, not just victimization or harassment. Past studies of online victimization and harassment often utilized variable-centered analyses in which online harassment or victimization predict a particular outcome. However, in this study, changes in individual participants overtime were examined by utilizing person-centered analyses. Specifically in this study I was able to examine the probability of participants designated in particular status at wave one transitioning to a status at wave two. This was particularly useful for analyses concerning the development of online harasser-victims over time. This study shows that harasser-victims at wave two are both victims at wave one who became harassers and harassers at wave one who later experienced victimization as well. However, very few 69 participants were uninvolved and became harasser-victims. This indicates that the harasser- victim group, which has the worst outcomes, may come to this detrimental online situation over time. It seems that high quality interventions could keep adolescents from transitioning to this group as it does not seem to be a sudden change for most adolescents as the transition form uninvolved to harasser-victim was rare. This study also examined the relationship between online peer support and victimization and/or harassment. When examining peer victimization online, it seems relevant to examine the nature of online friendships for adolescents. Despite the relationship between online communication and online harassment and victimization, there is little research that examines whether young people may find some online relationships to be beneficial or protective from negative online interactions. The findings from this study indicate that online peer support may be related to risky behaviors for some adolescents. However, it is important to note that higher online peer support at wave one was related to higher odds of harassers-only becoming uninvolved. These findings indicate that online peer support may be protective and that further longitudinal research is needed to understand costs and benefits of online peer support. This study also gives further evidence to the positive nature of social support from teachers, families, and offline peers, as these variables predicted greater odds of being uninvolved in online victimization at wave one. The findings concerning the possible protective nature of these types of social support was less clear, but overall, supportive relationships from online and offline peers, as well as families appear to be related to uninvolvement in online harassment and/or victimization. 70 Implications for Schools Interventions for in-school bullying have been implemented across schools in the United States, and now that online victimization is a problem for young people, schools are implementing interventions for online bullying and harassment (Mishna, Cook, Saini, Wu, & MacFadden, 2010). The findings from this study indicate that online harassment is somewhat malleable; however, helping young people avoid or escape on-going victimization may be more difficult. Schools and districts may have more immediate success with harassment cessation than helping young people escape victimization; however, it is still vital that students have the resources needed to end their victimization. Additionally, it seems that teachers may play an important but delicate role in online victimization and harassment interventions. Findings indicate that greater amounts of teacher support were related to lower odds of young people becoming harassers and slightly lower odds of becoming victims if uninvolved. However, higher teacher support for harassers-only was related to greater odds of harassers-only becoming harasser-victims and victims-only rather than staying in the harasser-only status. It seems that for some, high levels of teacher support may increase the chances of becoming a target of victimization. Therefore teachers and administrators may find better results by addressing the social climate of their classrooms to improve peer relationships rather than attempt to directly support or intervene with those involved in harassment of others. Further, a supportive school environment may create a space in which students can seek help from teachers for online victimization as research indicates that teachers are not often aware of online victimization experienced by students. 71 Conclusion The results from this study indicate that online victimization and harassment, though stable for many over the course of a year, can change and shift for adolescents. Examinations of the relationship among offline peer social support, teacher support, and family support indicate that those who report higher levels of these supports are more likely to be uninvolved. However, examinations of the relationship between support and changes over time revealed that the quality or type of support may be important. Though the findings of the study point to a multidimensional relationship among social support and future online harassment and victimization, information about the type and quality of social support from peers, teachers, and parents was not available in this dataset. Further, the small number of people at some transitions of interest precluded examinations of social support as a protective factor. Future work in this area should address these issues to better understand possible social support interventions for young people experiencing victimization or taking part in harassment online. 72 References Akaike, H. (1987). Factor analysis and AIC. Psychometrika, 52(3), 317-332. doi:10.1007/BF02294359 Aoyama, I., Saxon, T. F., & Fearon, D. D. (2011). Internalizing problems among cyberbullying victims and moderator effects of friendship quality. Multicultural Education & Technology Journal, 5(2), 92-105. Badaly, D., Kelly, B., Schwartz, D., & Dabney-Lieras, K. (2013). Longitudinal Associations of Electronic Aggression and Victimization with Social Standing During Adolescence. Journal of Youth and Adolescence, 42(6), 891-904. doi:10.1007/s10964-012-9787-2 Baldry, A., & Farrington, D. (2005). Protective Factors as Moderators of Risk Factors in Adolescence Bullying. Social Psychology of Education, 8(3), 263-284. doi:10.1007/s11218-005-5866-5 Bandura, A. (1978). Social Learning Theory of Aggression. Journal of Communication, 28(3), 12-29. doi:10.1111/j.1460-2466.1978.tb01621.x Bandura, A. (1986). Social foundations of thought and action: A social cognitive theory. Englewood Cliffs, NJ, US: Prentice-Hall, Inc. Bauman, S., Toomey, R. B., & Walker, J. L. (2013). Associations among bullying, cyberbullying, and suicide in high school students. Journal of Adolescence, 36(2), 341- 350. doi:http://dx.doi.org/10.1016/j.adolescence.2012.12.001 Beran, T., & Li, Q. (2007). The relationship between cyberbullying and school bullying. Journal of Student Wellbeing, 1(2), 15-33. Bergman, L. R., Magnusson, D., & El Khouri, B. M. (2003). Studying individual development in an interindividual context: A person-oriented approach: Psychology Press. 73 Bollmer, J. M., Milich, R., Harris, M. J., & Maras, M. A. (2005). A Friend in Need: The Role of Friendship Quality as a Protective Factor in Peer Victimization and Bullying. Journal of Interpersonal Violence, 20(6), 701-712. Bonanno, G. A. (2004). Loss, Trauma, and Human Resilience: Have We Underestimated the Human Capacity to Thrive After Extremely Aversive Events? American Psychologist, 59(1), 20-28. doi:10.1037/0003-066X.59.1.20 Bonanno, R. A., & Hymel, S. (2013). Cyber Bullying and Internalizing Difficulties: Above and Beyond the Impact of Traditional Forms of Bullying. Journal of Youth and Adolescence, 42(5), 685-697. doi:10.1007/s10964-013-9937-1 Bossler, A. M., Holt, T. J., & May, D. C. (2012). Predicting Online Harassment Victimization Among a Juvenile Population. Youth & Society, 44(4), 500-523. doi:10.1177/0044118x11407525 Bowes, L., Maughan, B., Caspi, A., Moffitt, T. E., & Arseneault, L. (2010). Families promote emotional and behavioural resilience to bullying: evidence of an environmental effect. Journal of Child Psychology and Psychiatry, 51(7), 809-817. doi:10.1111/j.1469- 7610.2010.02216.x Bradley, R. H., & Corwyn, R. F. (2002). Socioeconomic status and child development. Annual Review of Psychology, 53(371-399). doi:10.1146/annurev.psych.53.100901.135233 Burton, K. A., Florell, D., & Wygant, D. B. (2013). The role of peer attachment and normative beliefs about aggression on traditional bullying and cyberbullying. Psychology in the Schools, 50(2), 103-115. doi:10.1002/pits.21663 74 Bussey, K., & Bandura, A. (1999). Social cognitive theory of gender development and differentiation. Psychological Review, 106(4), 676-713. doi:10.1037/0033- 295x.106.4.676 Cappadocia, M. C., Craig, W. M., & Pepler, D. (2013). Cyberbullying: Prevalence, Stability, and Risk Factors During Adolescence. Canadian Journal of School Psychology, 28(2), 171- 192. Cassidy, W., Brown, K., & Jackson, M. (2012). ‘Under the radar’: Educators and cyberbullying in schools. School Psychology International, 33(5), 520-532. Chang, F.-C., Chiu, C.-H., Miao, N.-F., Chen, P.-H., Lee, C.-M., Huang, T.-F., & Pan, Y.-C. (2014). Online gaming and risks predict cyberbullying perpetration and victimization in adolescents. International Journal of Public Health, 1-10. doi:10.1007/s00038-014-0643- x Chang, F.-C., Lee, C.-M., Chiu, C.-H., Hsi, W.-Y., Huang, T.-F., & Pan, Y.-C. (2013). Relationships Among Cyberbullying, School Bullying, and Mental Health in Taiwanese Adolescents. The Journal of School Health, 83(6), 454. Christiansen, E. J., & Evans, W. P. (2005). Adolescent Victimization: Testing Models of Resiliency by Gender. The Journal of Early Adolescence, 25(3), 298-316. Collins, L. M., & Lanza, S. T. (2010). Latent Class and Latent Transition Analysis: With Applications in the Social, Behavioral, and Health Sciences by: Wiley. Conners ‐Burrow, N. A., Johnson, D. L., Whiteside ‐Mansell, L., McKelvey, L., & Gargus, R. A. (2009). Adults matter: Protecting children from the negative impacts of bullying. Psychology in the Schools, 46(7), 593-604. 75 Davidson, L. M., & Demaray, M. K. (2007). Social support as a moderator between victimization and internalizing-externalizing distress from bullying. School Psychology Review. Davis, K., & Koepke, L. (2015). Risk and protective factors associated with cyberbullying: Are relationships or rules more protective? Learning, Media and Technology, 1-25. doi:10.1080/17439884.2014.994219 Deater-Deckard, K., Dodge, K. A., Bates, J. E., & Pettit, G. S. (1998). Multiple risk factors in the development of externalizing behavior problems: Group and individual differences. Development and Psychopathology, 10(03), 469-493. Dempsey, A. G., Sulkowski, M. L., Dempsey, J., & Storch, E. A. (2011). Has cyber technology produced a new group of peer aggressors? Cyberpsychology, Behavior, and Social Networking, 14(5), 297-302. doi:10.1089/cyber.2010.0108 den Hamer, A. H., & Konijn, E. A. (2015). Adolescents' Media Exposure May Increase Their Cyberbullying Behavior: A Longitudinal Study. Journal of Adolescent Health, 56(2), 203-208. doi:http://dx.doi.org/10.1016/j.jadohealth.2014.09.016 DePaolis, K., & Williford, A. (2014). The Nature and Prevalence of Cyber Victimization Among Elementary School Children. Child & Youth Care Forum, 1-17. doi:10.1007/s10566-014- 9292-8 Donnon, T. (2010). Understanding How Resiliency Development Influences Adolescent Bullying and Victimization. Canadian Journal of School Psychology, 25(1), 101-113. Duggan, M., & Smith, A. (2013). Social media update 2013. Retrieved from <div data- angle="0" data-canvas-width="29.439799298100475" data-font-name="g_font_5_0" dir="ltr" style="font-size: 13.4pxRetrieved from left: 523.683pxRetrieved from top: 76 1196.87pxRetrieved from "> http://pewinternet.org/Reports/2013/Social-Media- Update.aspx Ellis, D. A., Zucker, R. A., & Fitzgerald, H. E. (1997). The role of family influences in development and risk. Alcohol Health & Research World, 21(3), 218-226. Espelage, D. L., Holt, M. K., & Henkel, R. R. (2003). Examination of Peer–Group Contextual Effects on Aggression During Early Adolescence. Child Development, 74(1), 205-220. doi:10.1111/1467-8624.00531 Evans, M., Kessler, D., Lewis, G., Peters, T. J., & Sharp, D. (2004). Assessing mental health in primary care research using standardized scales: can it be carried out over the telephone? Psychological Medicine, 34(01), 157-162. Evans, W. P., Marte, R. M., Betts, S., & Silliman, B. (2001). Adolescent Suicide Risk and Peer- Related Violent Behaviors and Victimization. Journal of Interpersonal Violence, 16(12), 1330-1348. doi:10.1177/088626001016012006 Fanti, K. A., Demetriou, A. G., & Hawa, V. V. (2012). A longitudinal study of cyberbullying: Examining risk and protective factors. European Journal of Developmental Psychology, 9(2), 168-181. doi:10.1080/17405629.2011.643169 Fanti, K. A., & Kimonis, E. R. (2013). Dimensions of Juvenile Psychopathy Distinguish "Bullies," "Bully-Victims," and "Victims". Psychology of Violence, 3(4), 396-409. doi:10.1037/a0033951 Farrington, D. P., Loeber, R., & Ttofi, M. M. (2011). Risk and protective factors. In B. C. Welsh & D. P. Farrington (Eds.), The Oxford handbook of crime prevention. New York, NY: Oxford University Press. 77 Fergus, S., & Zimmerman, M. A. (2005). Adolescent Resilience: A framework for understanding healthy development in the face of risk. Annual Review of Public Health, 26, 399-419. Festl, R., Scharkow, M., & Quandt, T. (2014). The Individual or the Group: A Multilevel Analysis of Cyberbullying in School Classes. Human Communication Research, n/a-n/a. doi:10.1111/hcre.12056 Finkelhor, D., Mitchell, K., & Wolak, J. (2000). Online Victimization: A Report on the Nation's Youth. Retrieved from Flaspohler, P. D. E., Jennifer L. Vanderzee, Karin L. Sink, Holli E. Birchmeier, Zachary. (2009). Stand by me: The effects of peer and teacher support in mitigating the impact of bullying on quality of life. Psychology in the Schools, 46(7), 636–649. Forsdahl, A. (1977). Are poor living conditions in childhood and adolescence an important risk factor for arteriosclerotic heart disease? British Journal of Preventive & Social Medicine, 31(2), 91-95. Garmezy, N. (1985). Stress-resistant children: The search for protective factors. Recent research in developmental psychopathology, 4, 213-233. Garmezy, N., Masten, A. S., & Tellegen, A. (1984). The Study of Stress and Competence in Children: A Building Block for Developmental Psychopathology. Child Development, 55(1), 97-111. doi:10.2307/1129837 Gradinger, P., Strohmeier, D., Schiller, E. M., Stefanek, E., & Spiel, C. (2012). Cyber- victimization and popularity in early adolescence: Stability and predictive associations. European Journal of Developmental Psychology, 9(2), 228-243. doi:10.1080/17405629.2011.643171 78 Graham, J., Olchowski, A., & Gilreath, T. (2007). How Many Imputations are Really Needed? Some Practical Clarifications of Multiple Imputation Theory. Prevention Science, 8(3), 206-213. doi:10.1007/s11121-007-0070-9 Graham, S., & Juvonen, J. (1998). A social cognitive perspective on peer aggression and victimizaiton. Annals of Child Development, 13, 21-66. Guynn, J., & Stobart, J. (2013, 2013 Aug 20). New social site, same bullying; Ask.fm, popular with teens, has been linked to five suicides and says it will take steps to deter harassment. Los Angeles Times, p. A1. Retrieved from http://search.proquest.com.libproxy.usc.edu/latimes/docview/1426205692/6E185799707 A432DPQ/15?accountid=14749 Gámez-Guadix, M., Orue, I., Smith, P. K., & Calvete, E. (2013). Longitudinal and Reciprocal Relations of Cyberbullying With Depression, Substance Use, and Problematic Internet Use Among Adolescents. Journal of Adolescent Health, 53(4), 446-452. doi:http://dx.doi.org/10.1016/j.jadohealth.2013.03.030 Haine, R. A., Ayers, T. S., Sandler, I. N., Wolchik, S. A., & Weyer, J. L. (2003). Locus of control and self-esteem as stress-moderators or stress-mediators in parentally bereaved children. Death Studies, 27(7), 619-640. doi:10.1080/07481180302894 Haynie, D. L., Nansel, T., Eitel, P., Crump, A. D., Saylor, K., Yu, K., & Simons-Morton, B. (2001). Bullies, Victims, and Bully/Victims:: Distinct Groups of At-Risk Youth. The Journal of Early Adolescence, 21(1), 29-49. Hemphill, S. A., & Heerde, J. A. (2014). Adolescent Predictors of Young Adult Cyberbullying Perpetration and Victimization Among Australian Youth. Journal of Adolescent Health, 55(4), 580-587. doi:http://dx.doi.org/10.1016/j.jadohealth.2014.04.014 79 Hemphill, S. A., Kotevski, A., & Heerde, J. A. (2015). Longitudinal associations between cyber- bullying perpetration and victimization and problem behavior and mental health problems in young Australians. International Journal of Public Health, 60(2), 227-237. doi:10.1007/s00038-014-0644-9 Hemphill, S. A., Kotevski, A., Tollit, M., Smith, R., Herrenkohl, T. I., Toumbourou, J. W., & Catalano, R. F. (2012). Longitudinal Predictors of Cyber and Traditional Bullying Perpetration in Australian Secondary School Students. Journal of Adolescent Health, 51(1), 59-65. doi:http://dx.doi.org/10.1016/j.jadohealth.2011.11.019 Hemphill, S. A., Tollit, M., & Herrenkohl, T. I. (2013). Protective Factors Against the Impact of School Bullying Perpetration and Victimization on Young Adult Externalizing and Internalizing Problems. Journal of School Violence, 13(1), 125-145. doi:10.1080/15388220.2013.844072 Hemphill, S. A., Tollit, M., Kotevski, A., & Heerde, J. A. (2014). Predictors of Traditional and Cyber-Bullying Victimization: A Longitudinal Study of Australian Secondary School Students. Journal of Interpersonal Violence. Hillier, L., Mitchell, K. J., & Ybarra, M. L. (2012). The Internet As a Safety Net: Findings From a Series of Online Focus Groups With LGB and Non-LGB Young People in the United States. Journal of LGBT Youth, 9(3), 225-246. doi:10.1080/19361653.2012.684642 Hinduja, S., & Patchin, J. W. (2007). Offline Consequences of Online Victimization. Journal of School Violence, 6(3), 89-112. doi:10.1300/J202v06n03_06 Hinduja, S., & Patchin, J. W. (2008). Cyberbullying: An Exploratory Analysis of Factors Related to Offending and Victimization. Deviant Behavior, 29(2), 129-156. doi:10.1080/01639620701457816 80 Holt, M. K., & Espelage, D. L. (2007). Perceived Social Support among Bullies, Victims, and Bully-Victims. Journal of Youth and Adolescence, 36(8), 984-994. doi:10.1007/s10964- 006-9153-3 Hugh-Jones, S., & Smith, P. K. (1999). Self-reports of short- and long-term effects of bullying on children who stammer. British Journal of Educational Psychology, 69(2), 141-158. doi:10.1348/000709999157626 Jones, L. M., Mitchell, K. J., & Finkelhor, D. (2012). Trends in Youth Internet Victimization: Findings From Three Youth Internet Safety Surveys 2000–2010. Journal of Adolescent Health, 50(2), 179-186. doi:10.1016/j.jadohealth.2011.09.015 Jose, P. E., Kljakovic, M., Scheib, E., & Notter, O. (2012). The Joint Development of Traditional Bullying and Victimization With Cyber Bullying and Victimization in Adolescence. Journal of Research on Adolescence, 22(2), 301-309. doi:10.1111/j.1532- 7795.2011.00764.x Juvonen, J., & Gross, E. F. (2008). Extending the School Grounds?—Bullying Experiences in Cyberspace. Journal of School Health, 78(9), 496-505. doi:10.1111/j.1746- 1561.2008.00335.x Kaltiala-Heino, R., Rimpelä, M., Rantanen, P., & Rimpelä, A. (2000). Bullying at school—an indicator of adolescents at risk for mental disorders. Journal of Adolescence, 23(6), 661- 674. doi:http://dx.doi.org/10.1006/jado.2000.0351 Klomek, A. B., Marrocco, F., Kleinman, M., Schonfeld, I., & Gould, M. (2008). Peer Victimization, Depression, and Suicidiality in Adolescents. Suicide & Life - Threatening Behavior, 38(2), 166. 81 Klomek, A. B., Sourander, A. M. D., & Gould, M. P. M. P. H. (2010). The Association of Suicide and Bullying in Childhood to Young Adulthood: A Review of Cross-Sectional and Longitudinal Research Findings. Canadian Journal of Psychiatry, 55(5), 282-288. Konishi, C., Hymel, S., Zumbo, B. D., & Zhen, L. (2010). Do School Bullying and Student— Teacher Relationships Matter for Academic Achievement? A Multilevel Analysis. Canadian Journal of School Psychology, 25(1), 19-39. Kowalski, R. M., Giumetti, G. W., Schroeder, A. N., & Lattanner, M. R. (2014). Bulling in the Digital Age: A Critical Review and Meta-Analysis of Cyberbullying Research Among Youth. Psychological Bulletin(Advance online publication.). Kowalski, R. M., Limber, S., & Agatston, P. W. (2008). Cyber bullying: Bullying in the digital age. Malden, MA: Blackwell. Kowalski, R. M., Limber, S., & Agatston, P. W. (2012). Cyberbullying: Bullying in the digital age (2nd ed.). Malden, MA: Blackwell. Kowalski, R. M., & Limber, S. P. (2007). Electronic Bullying Among Middle School Students. Journal of Adolescent Health, 41(6, Supplement), S22-S30. doi:10.1016/j.jadohealth.2007.08.017 Kowalski, R. M., & Limber, S. P. (2013). Psychological, Physical, and Academic Correlates of Cyberbullying and Traditional Bullying. Journal of Adolescent Health, 53(1, Supplement), S13-S20. doi:http://dx.doi.org/10.1016/j.jadohealth.2012.09.018 Langos, C. (2012). Cyberbullying: The Challenge to Define. Cyberpsychology Behavior and Social Networking, 15(6), 285-289. doi:10.1089/cyber.2011.0588 82 Law, D. M., Shapka, J. D., & Olson, B. F. (2010). To control or not to control? Parenting behaviours and adolescent online aggression. Computers in Human Behavior, 26(6), 1651-1656. doi:http://dx.doi.org/10.1016/j.chb.2010.06.013 Lenhart, A., Madden, M., Smith, A., Purcell, K., Kathryn, Z., & Rainie, L. (Producer). (2011, April 22, 2015). Teens, kindness and cruelty on social network sites. Retrieved from http://www.pewinternet.org/2011/11/09/teens-kindness-and-cruelty-on-social-network- sites/ Levin, E. K., Zigmond, N., & Birch, J. W. (1985). A follow-up study of 52 learning disabled adolescents. Journal of Learning Disabilities, 18(1), 2-7. Li, Q. (2006). Cyberbullying in Schools: A Research of Gender Differences. School Psychology International, 27(2), 157-170. doi:10.1177/0143034306064547 Li, Q. (2007). New bottle but old wine: A research of cyberbullying in schools. Computers in Human Behavior, 23(4), 1777-1791. doi:http://dx.doi.org/10.1016/j.chb.2005.10.005 Livingstone, S., & Smith, P. K. (2014). Annual Research Review: Harms experienced by child users of online and mobile technologies: the nature, prevalence and management of sexual and aggressive risks in the digital age. Journal of Child Psychology and Psychiatry, n/a-n/a. doi:10.1111/jcpp.12197 Luthar, S. S., Cicchetti, D., & Becker, B. (2000). The Construct of Resilience: A Critical Evaluation and Guidelines for Future Work. Child Development, 71(3), 543-562. doi:10.1111/1467-8624.00164 Ma, X. (2001). Bullying and Being Bullied: To What Extent Are Bullies Also Victims? American Educational Research Journal, 38(2), 351-370. 83 Machmutow, K., Perren, S., Sticca, F., & Alsaker, F. D. (2012). Peer victimisation and depressive symptoms: can specific coping strategies buffer the negative impact of cybervictimisation? Emotional and Behavioural Difficulties, 17(3-4), 403-420. doi:10.1080/13632752.2012.704310 Malecki, C. K., & Demaray, M. K. (2002). Measuring perceived social support: Development of the child and adolescent social support scale (CASSS). Psychology in the Schools, 39(1), 1-18. doi:10.1002/pits.10004 Masten, A. S., Best, K. M., & Garmezy, N. (1990). Resilience and development: Contributions from the study of children who overcome adversity. Development and Psychopathology, 2(04), 425-444 M423 - 410.1017/S0954579400005812. Masten, A. S., & Garmezy, N. (1985). Risk, Vulnerability, and Protective Factors in Developmental Psychopathology. In B. Lahey & A. Kazdin (Eds.), Advances in Clinical Child Psychology (Vol. 8, pp. 1-52): Springer US. Masten, A. S., & Powell, J. L. (2003). A resilience framework for research, policy, and practice. In S. S. Luthar (Ed.), Resilience and vulnerability: Adaptation in the context of childhood adversities (pp. 1-28). Cambridge, UK: Cambridge University Press. McArdle, J. J. (2013). Dealing with Longitudinal Attrition Using Logistic Regression and Decision Tree Analyses. In J. J. McArdle & G. Ritschard (Eds.), Contemporary issues in exploratory data mining in the behavioral sciences (pp. 282-311). New York, NY: Taylor & Francis. Menesini, E., Nocentini, A., Palladino, B. E., Frisén, A., Berne, S., Ortega-Ruiz, R., . . . Smith, P. K. (2012). Cyberbullying Definition Among Adolescents: A Comparison Across Six 84 European Countries. CyberPsychology, Behavior & Social Networking, 15(9), 455-463. doi:10.1089/cyber.2012.0040 Mishna, F., Cook, C., Gadalla, T., Daciuk, J., & Solomon, S. (2010). Cyber Bullying Behaviors Among Middle and High School Students. American Journal of Orthopsychiatry, 80(3), 362-374. doi:10.1111/j.1939-0025.2010.01040.x Mishna, F., Cook, C., Saini, M., Wu, M.-J., & MacFadden, R. (2010). Interventions to Prevent and Reduce Cyber Abuse of Youth: A Systematic Review. Research on Social Work Practice. Mishna, F., Khoury-Kassabri, M., Gadalla, T., & Daciuk, J. (2012). Risk factors for involvement in cyber bullying: Victims, bullies and bully-victims. Children and Youth Services Review, 34(1), 63-70. doi:10.1016/j.childyouth.2011.08.032 Mitchell, K. J., Finkelhor, D., Wolak, J., Ybarra, M. L., & Turner, H. (2011). Youth Internet Victimization in a Broader Victimization Context. Journal of Adolescent Health, 48(2), 128-134. doi:10.1016/j.jadohealth.2010.06.009 Mitchell, K. J., Ybarra, M., & Finkelhor, D. (2007). The Relative Importance of Online Victimization in Understanding Depression, Delinquency, and Substance Use. Child Maltreatment, 12(4), 314-324. doi:10.1177/1077559507305996 Modecki, K., Barber, B., & Vernon, L. (2013). Mapping Developmental Precursors of Cyber- Aggression: Trajectories of Risk Predict Perpetration and Victimization. Journal of Youth and Adolescence, 42(5), 651-661. doi:10.1007/s10964-012-9887-z Moore, P., Huebner, E., & Hills, K. (2012). Electronic Bullying and Victimization and Life Satisfaction in Middle School Students. Social Indicators Research, 107(3), 429-447. doi:10.1007/s11205-011-9856-z 85 Moore, R., Guntupalli, N. T., & Lee, T. (2010). Parental regulation and online activities: Examining factors that influence a youth's potential to become a victim of online harassment. International Journal of Cyber Criminology, 4(1-2), 685-698. Morrison, G. M., & Cosden, M. A. (1997). Risk, Resilience, and Adjustment of Individuals with Learning Disabilities. Learning Disability Quarterly, 20(1), 43-60. doi:10.2307/1511092 Mossakowski, K. N. (2003). Coping with perceived discrimination: Does ethnic identity protect mental health?*. Journal of Health and Social Behavior, 44(3), 318-331. doi:10.1111/j.1471-6402.1987.tb00921.x Muthén, L. K., & Muthén, B. O. (1998-2010). Mplus User’s Guide. Sixth Edition. Los Angeles, CA: Muthén & Muthén. Navarro, R., Yubero, S., Larrañaga, E., & Martínez, V. (2012). Children’s Cyberbullying Victimization: Associations with Social Anxiety and Social Competence in a Spanish Sample. Child Indicators Research, 5(2), 281-295. doi:10.1007/s12187-011-9132-4 O'Brennan, L. M., Bradshaw, C. P., & Sawyer, A. L. (2009). Examining Developmental Differences In The Social-Emotional Problems Among Frequent Bullies, Victims, And Bully/Victims. Psychology in the Schools, 46(2), 100-115. doi:10.1002/pits.20357 Olweus, D. (1993). Bullying at school: what we know and what we can do. New York: Blackwell. Patchin, J. W., & Hinduja, S. (2006). Bullies Move Beyond the Schoolyard : A Preliminary Look at Cyberbullying. Youth Violence and Juvenile Justice, 4(2), 148-169. doi:10.1177/1541204006286288 Patchin, J. W., & Hinduja, S. (2010). Cyberbullying and Self-Esteem*. Journal of School Health, 80(12), 614-621. doi:10.1111/j.1746-1561.2010.00548.x 86 Patel, V., & Kleinman, A. (2003). Poverty and common mental disorders in developing countries. Bulletin of The World Health Organization, 81(8), 609-615. Pearce, M. (2013, 2013 Sep 13). Officials cite cyberbullying in girl's death; Former friends 'beat down' Rebecca Ann Sedwick, 12, online before her apparent suicide in Florida, investigators say. Los Angeles Times. Retrieved from http://search.proquest.com.libproxy.usc.edu/latimes/docview/1432047782/fulltext/6E185 799707A432DPQ/1?accountid=14749 Perry, D. G., Perry, L. C., & Rasmussen, P. (1986). Cognitive Social Learning Mediators of Aggression. Child Development, 57(3), 700-711. doi:10.2307/1130347 Pfeil, U., Arjan, R., & Zaphiris, P. (2009). Age differences in online social networking – A study of user profiles and the social capital divide among teenagers and older users in MySpace. Computers in Human Behavior, 25(3), 643-654. doi:http://dx.doi.org/10.1016/j.chb.2008.08.015 Ragan, P. V., & McGlashan, T. H. (1986). Childhood parental death and adult psychopathology. The American Journal of Psychiatry, 143(2), 153-157. Raskauskas, J. L., Gregory, J., Harvey, S. T., Rifshana, F., & Evans, I. M. (2010). Bullying among primary school children in New Zealand: relationships with prosocial behaviour and classroom climate. Educational Research, 52(1), 1-13. doi:10.1080/00131881003588097 Roberto, A. J., Eden, J., Savage, M. W., Ramos-Salazar, L., & Deiss, D. M. (2014). Prevalence and Predictors of Cyberbullying Perpetration by High School Seniors. Communication Quarterly, 62(1), 97-114. doi:10.1080/01463373.2013.860906 87 Rose, C. A., & Tynes, B. M. Longitudinal Associations Between Cybervictimization and Mental Health Among U.S. Adolescents. Journal of Adolescent Health. doi:http://dx.doi.org/10.1016/j.jadohealth.2015.05.002 Rose, C. A., & Tynes, B. M. (2015). Longitudinal Associations Between Cybervictimization and Mental Health Among U.S. Adolescents. Journal of Adolescent Health. doi:http://dx.doi.org/10.1016/j.jadohealth.2015.05.002 Rose, C. A., & Tynes, B. M. (in press). Longitudinal Associations Between Cybervictimization and Mental Health Among U.S. Adolescents. Journal of Adolescent Health. doi:http://dx.doi.org/10.1016/j.jadohealth.2015.05.002 Rosenthal, S., Feiring, C., & Taska, L. (2003). Emotional support and adjustment over a year’s time following sexual abuse discovery. Child Abuse & Neglect, 27(6), 641-661. doi:http://dx.doi.org/10.1016/S0145-2134(03)00104-2 Rutter, M. (1979). Protective factors in children's responses to stress and disadvantage. Ann Acad Med Singapore, 8(3), 324-338. Rutter, M. (1985). Resilience in the face of adversity: Protective factors and resistance to psychiatric disorder. The British Journal of Psychiatry, 147, 598-611. doi:10.1192/bjp.147.6.598 Schafer, J. L., & Graham, J. W. (2002). Missing data: Our view of the state of the art. Psychological Methods, 7(2), 147-177. doi:http://dx.doi.org/10.1037/1082-989X.7.2.147 Schmidt, M. E., & Bagwell, C. L. (2007). The protective role of friendships in overtly and relationally victimized boys and girls. Merrill-Palmer Quarterly, 53(3), 439-460. Schwarz, G. (1978). Estimating the Dimension of a Model. 461-464. doi:10.1214/aos/1176344136 88 Smith, P. K., Mahdavi, J., Carvalho, M., Fisher, S., Russell, S., & Tippett, N. (2008). Cyberbullying: its nature and impact in secondary school pupils. Journal of Child Psychology and Psychiatry, 49(4), 376-385. doi:10.1111/j.1469-7610.2007.01846.x Sourander, A., Brunstein Klomek, A., Ikonen, M., Lindroos, J., Luntamo, T., Koskelainen, M., . . . Helenius, H. (2010). Psychosocial risk factors associated with cyberbullying among adolescents: a population-based study. Archives of general psychiatry, 67(7), 720-728. doi:10.1001/archgenpsychiatry.2010.79 Spears, B., Taddeo, C., Daly, A., Stretton, A., & Karklins, L. (2015). Cyberbullying, help- seeking and mental health in young Australians: implications for public health. International Journal of Public Health, 60(2), 219-226. doi:10.1007/s00038-014-0642-y StataCorp. (2011). Stata Statistical Software: Release 12. College Station, TX: StataCorp LP. Steffgen, G., König, A., Pfetsch, J., & Melzer, A. (2011). Are Cyberbullies Less Empathic? Adolescents' Cyberbullying Behavior and Empathic Responsiveness. Cyberpsychology, Behavior, and Social Networking, 14(11), 643-648. doi:doi:10.1089/cyber.2010.0445 Sticca, F., Ruggieri, S., Alsaker, F., & Perren, S. (2013). Longitudinal Risk Factors for Cyberbullying in Adolescence. Journal of Community & Applied Social Psychology, 23(1), 52-67. doi:10.1002/casp.2136 Toblin, R. L., Schwartz, D., Hopmeyer Gorman, A., & Abou-ezzeddine, T. (2005). Social– cognitive and behavioral attributes of aggressive victims of bullying. Journal of Applied Developmental Psychology, 26(3), 329-346. doi:http://dx.doi.org/10.1016/j.appdev.2005.02.004 89 Tokunaga, R. S. (2010). Following you home from school: A critical review and synthesis of research on cyberbullying victimization. Computers in Human Behavior, 26(3), 277-287. doi:10.1016/j.chb.2009.11.014 Tu, K. M., Erath, S. A., & Flanagan, K. S. (2012). Can socially adept friends protect peer- victimized early adolescents against lower academic competence? Journal of Applied Developmental Psychology, 33(1), 24-30. doi:http://dx.doi.org/10.1016/j.appdev.2011.09.002 Tynes, B. M., Giang, M. T., Williams, D. R., & Thompson, G. N. (2008). Online Racial Discrimination and Psychological Adjustment Among Adolescents. Journal of Adolescent Health, 43(6), 565-569. doi:10.1016/j.jadohealth.2008.08.021 Tynes, B. M., Reynolds, L., & Greenfield, P. M. (2004). Adolescence, race, and ethnicity on the Internet: A comparison of discourse in monitored vs. unmonitored chat rooms. Journal of Applied Developmental Psychology, 25(6), 667-684. doi:http://dx.doi.org/10.1016/j.appdev.2004.09.003 Tynes, B. M., Rose, C. A., & Williams, D. R. (2010). The development and validation of the online victimization scale for adolescents. Cyberpsychology: Journal of Psychosocial Research on Cyberspace, 4(2), 1-15. Tynes, B. M., Umaña-Taylor, A. J., Rose, C. A., Lin, J., & Anderson, C. J. (2012). Online racial discrimination and the protective function of ethnic identity and self-esteem for African American adolescents. Developmental Psychology, 48(2), 343. Valkenburg, P. M., Schouten, A. P., & Peter, J. (2005). Adolescents’ identity experiments on the internet. New Media & Society, 7(3), 383-402. 90 Varjas, K., Henrich, C. C., & Meyers, J. (2009). Urban Middle School Students' Perceptions of Bullying, Cyberbullying, and School Safety. Journal of School Violence, 8(2), 159-176. doi:10.1080/15388220802074165 Wachs, S. (2012). Moral disengagement and emotional and social difficulties in bullying and cyberbullying: differences by participant role. Emotional and behavioural difficulties, 17(3-4), 347-360. Wade, A., & Beran, T. (2011). Cyberbullying: The New Era of Bullying. Canadian Journal of School Psychology, 26(1), 44-61. Walrave, M., & Heirman, W. (2011). Cyberbullying: Predicting Victimisation and Perpetration. Children & Society, 25(1), 59-72. doi:10.1111/j.1099-0860.2009.00260.x Wang, J., Iannotti, R. J., & Nansel, T. R. (2009). School Bullying Among Adolescents in the United States: Physical, Verbal, Relational, and Cyber. Journal of Adolescent Health, 45(4), 368-375. doi:10.1016/j.jadohealth.2009.03.021 Wang, J., Nansel, T. R., & Iannotti, R. J. (2011). Cyber and Traditional Bullying: Differential Association With Depression. Journal of Adolescent Health, 48(4), 415-417. doi:http://dx.doi.org/10.1016/j.jadohealth.2010.07.012 Wegge, D., Vandebosch, H., Eggermont, S., & Pabian, S. (2014). Popularity Through Online Harm: The Longitudinal Associations Between Cyberbullying and Sociometric Status in Early Adolescence. The Journal of Early Adolescence. Werner, N., Bumpus, M., & Rock, D. (2010). Involvement in Internet Aggression During Early Adolescence. Journal of Youth and Adolescence, 39(6), 607-619. doi:10.1007/s10964- 009-9419-7 91 Williams, K. R., & Guerra, N. G. (2007). Prevalence and Predictors of Internet Bullying. Journal of Adolescent Health, 41(6, Supplement), S14-S21. doi:10.1016/j.jadohealth.2007.08.018 Williams, N. R., Davey, M., & Klock-Powell, K. (2003). Rising from the Ashes. Social Work in Health Care, 36(4), 53-77. doi:10.1300/J010v36n04_04 Wolak, J., Mitchell, K. J., & Finkelhor, D. (2007). Does online harassment constitute bullying? An exploration of online harassment by known peers and online-only contacts. The Journal of adolescent health : official publication of the Society for Adolescent Medicine, 41(6 Suppl 1), S51-S58. doi:10.1016/j.jadohealth.2007.08.019 Wright, M. F., & Li, Y. (2012). Kicking the digital dog: a longitudinal investigation of young adults' victimization and cyber-displaced aggression. Cyberpsychol Behav Soc Netw, 15(9), 448-454. doi:10.1089/cyber.2012.0061 Yang, S.-J., Stewart, R., Kim, J.-M., Kim, S.-W., Shin, I.-S., Dewey, M., . . . Yoon, J.-S. (2013). Differences in predictors of traditional and cyber-bullying: a 2-year longitudinal study in Korean school children. European Child & Adolescent Psychiatry, 22(5), 309-318. doi:10.1007/s00787-012-0374-6 Ybarra, M. L. (2004). Linkages between depressive symptomatology and Internet harassment among young regular Internet users. Cyberpsychology & Behavior, 7(2), 247-257. doi:10.1089/109493104323024500 Ybarra, M. L., Boyd, D., Korchmaros, J. D., & Oppenheim, J. (2012). Defining and Measuring Cyberbullying Within the Larger Context of Bullying Victimization. Journal of Adolescent Health, 51(1), 53-58. doi:10.1016/j.jadohealth.2011.12.031 92 Ybarra, M. L., Diener-West, M., & Leaf, P. J. (2007). Examining the overlap in Internet harassment and school bullying: Implications for school intervention. Journal of Adolescent Health, 41(6), S42-S50. doi:10.1016/j.jadohealth.2007.09.004 Ybarra, M. L., & Mitchell, K. J. (2004). Online aggressor/targets, aggressors, and targets: a comparison of associated youth characteristics. Journal of Child Psychology and Psychiatry, 45(7), 1308-1316. doi:10.1111/j.1469-7610.2004.00328.x Ybarra, M. L., & Mitchell, K. J. (2008). How risky are social networking sites? A comparison of places online where youth sexual solicitation and harassment occurs. Pediatrics, 121(2), e350-357. Ybarra, M. L., Mitchell, K. J., Wolak, J., & Finkelhor, D. (2006). Examining Characteristics and Associated Distress Related to Internet Harassment: Findings From the Second Youth Internet Safety Survey. Pediatrics, 118(4), e1169-e1177. Zhang, H., Burton. (1999). Recursive Partitioning in the Health Sciences. New York: Springer. Zimet, G. D., Dahlem, N. W., Zimet, S. G., & Farley, G. K. (1988). The Multidimensional Scale of Perceived Social Support. Journal of Personality Assessment, 52(1), 30-41. doi:10.1207/s15327752jpa5201_2 . 93 TABLES Table 3.1 Items Used in LTA and LCA Models Victim 1 People have posted mean or rude things about me on the Internet. Victim 2 I have been harassed or bothered online because of something that happened at school. Harass 1 Because someone called you a name Harass 2 Because something that happened at school Harass 3 Because of the way the person acts 94 Table 4.1 Means and Standard Deviations For All Variables Time 1 Time 2 Time 3 Variable M (SD) M (SD) M (SD) Min Max Victim 1 0.16 (0.37) 0.35 (0.48) 0.40 (0.49) 0 1 Victim 2 0.11 (0.31) 0.24 (0.42) 0.30 (0.46) 0 1 Harass 1 0.35 (0.48) 0.33 (0.47) 0.30 (0.46) 0 1 Harass2 0.19 (0.39) 0.16 (0.37) 0.12 (0.32) 0 1 Harass3 0.22 (0.42) 0.23 (0.42) 0.22 (0.42) 0 1 Peer Support Offline 5.39 (1.48) 5.30 (1.48) 5.40 (1.40) 1 7 Family Support 5.36 (1.52) 5.21 (1.48) 5.30 (1.44) 1 7 Teacher Support 5.02 (1.35) 4.89 (1.36) 4.96 (1.30) 1 7 Peer Support Online 2.27 (1.22) 2.25 (1.25) 2.23 (1.20) 1 5 Note. All Victim and Harass variables were coded 0 and 1. All support Variables are unstandardized 95 Table 4.2 Correlation Matrix for All Variables 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 1. Victim 1 T1 1 2. Victim 2 T1 0.53 1 3. Harass 1 T1 0.24 0.17 1 4. Harass 2 T1 0.25 0.23 0.39 1 5. Harass 3 T1 0.30 0.16 0.36 0.39 1 6. Victim 1 T2 0.27 0.24 0.17 0.23 0.17 1 7. Victim 2 T2 0.24 0.27 0.16 0.15 0.13 0.51 1 8. Harass 1 T2 0.15 0.11 0.33 0.21 0.15 0.20 0.20 1 9. Harass 2 T2 0.06 0.11 0.14 0.17 0.11 0.23 0.18 0.42 1 10. Harass 3 T2 0.14 0.07 0.14 0.16 0.28 0.23 0.19 0.40 0.35 1 11. Victim 1 T3 0.24 0.19 0.09 0.12 0.13 0.40 0.26 0.19 0.11 0.23 1 12. Victim 2 T3 0.17 0.13 0.05 0.05 0.09 0.34 0.30 0.12 0.16 0.16 0.53 1 13. Harass 1 T3 0.19 0.16 0.26 0.19 0.19 0.23 0.20 0.39 0.27 0.32 0.34 0.25 1 14. Harass 2 T3 0.09 0.22 0.18 0.15 0.12 0.21 0.23 0.20 0.29 0.19 0.23 0.27 0.36 1 15. Harass 3 T3 0.05 0.12 0.11 0.12 0.15 0.24 0.12 0.19 0.15 0.35 0.22 0.17 0.43 0.30 1 16. Teacher Support -0.16 -0.11 -0.16 -0.12 -0.12 -0.17 -0.16 -0.16 -0.04 -0.12 -0.10 -0.07 -0.05 -0.05 -0.10 1 17. Family Support -0.17 -0.11 -0.11 -0.08 -0.13 -0.13 -0.14 -0.12 -0.03 -0.09 -0.11 -0.04 -0.06 -0.01 0.04 0.66 1 18. Offline Friend Support -0.08 -0.06 -0.08 -0.02 -0.04 0.00 -0.03 -0.18 -0.03 -0.06 -0.03 0.03 -0.06 0.03 0.04 0.54 0.55 1 19. Online Friend Support 0.18 0.18 0.13 0.10 0.12 0.15 0.19 0.07 0.12 0.09 0.14 0.07 0.08 0.12 0.11 -0.04 -0.05 0.14 96 Table 4.3 Fit indices for LCA models with 3 –5 Classes for Each Wave Number of Classes 2 3 4 5 Wave 1 AIC 4510.379 4423.028 4419.142 4415.172 BIC 4564.689 4506.963 4532.701 4558.354 Wave 2 AIC 5279.134 5132.75 5107.789 5116.83 BIC 5333.444 5216.684 5221.347 5260.012 Wave 3 AIC 5243.93 5127.014 5096.069 5109.598 BIC 5298.24 5210.948 5209.627 5252.781 97 Table 4.4 Items Response Probabilities and Status Prevalences at Waves One and Two Latent Statuses Harasser- Victim Victim- Only Harasser- Only Uninvolved Wave 1 status prevalences (%) 90.5 (8.8%) 54.5 (5.3%) 288.6 (28.0%) 596.4 (57.9%) Wave 2 status prevalences (%) 144.9 (14.1%) 246.2 (23.9%) 145.6 (14.1%) 493.3 (47.9%) Item-response probabilities of endorsement V1. People posted rude things about me 0.862 0.896 0.078 0.029 V2.Was harassed online because of something at school 0.574 0.613 0.059 0.024 H1. Harassed others because someone called you a name 0.852 0.275 0.73 0.096 H2. Harassed others because something that happened at school 0.763 0.043 0.401 0.012 H3. Harassed others because of the way the person acts 0.688 0.21 0.47 0.037 Note. Item-response probabilities held constant over time, therefore probabilities are the same for waves one and two. 98 Table 4.5 Transition Probabilities Table Wave 2 Status H-V V H U Wave 1 Status H-V 0.536 0.335 0.037 0.092 V 0.132 0.811 0.002 0.055 H 0.212 0.267 0.361 0.161 U 0.047 0.159 0.064 0.730 99 Table 4.6 Logistic Regression Results For Covariates Predicting Wave One Status Latent Status Time 1 Harasser-Victim Harasser Victim Offline Peer Support (-2LL=26.012, df=3, p < 0.001) β -0.198 -0.392 -0.606 Odds Ratio 0.820 0.676 0.546 Latent Status Time 1 Harasser-Victim Harasser Victim Online Peer Support (-2LL=51.434, df=3, p < 0.001) β 0.665 0.803 0.312 Odds Ratio 1.944 2.232 1.366 Latent Status Time 1 Harasser-Victim Harasser Victim Teacher Support (-2LL=73.426, df=3, p < 0.001) β -0.489 -0.669 -1.071 Odds Ratio 0.613 0.512 0.343 Latent Status Time 1 Harasser-Victim Harasser Victim Family Support (-2LL=56.842, df=3, p < 0.001) β -0.576 -0.895 -0.472 Odds Ratio 0.562 0.409 0.624 Note. For all analyses in this table, the uninvolved group is the reference 100 Table 4.7 Count of Participants for Each Transition Wave 2 Status H-V V H U Wave 1 Status H-V 48.6 30.3 3.4 8.3 V 7.1 44.2 0.1 3.0 H 61.1 77.0 104.1 46.4 U 28.1 94.8 37.9 435.6 Note. Counts are averaged across 20 imputed datasets 101 Table 4.7 Odds Ratios for Social Support Variables Predicting Transitions from Wave One Statuses to Wave Two Statuses Time 2 Latent Status Time 1 Latent Status Harasser-Victim Victim Harasser Uninvolved Online Peer Support (-2LL=59.864, df=11, p < 0.001) Harasser-Victim -- 0.912 0.024† 1.474† Victim 0.180† -- *** 0.608† Harasser 1.725 1.260 -- 1.455 Uninvolved 1.162 1.133 1.250 -- Offline Peer Support (-2LL=46.966, df=11, p < 0.001) Harasser-Victim -- 1.001 167.670† 1.543† Victim 0.757† -- *** 0.307† Harasser 1.435 1.125 -- 1.490 Uninvolved 0.645 1.028 0.525 -- Teacher Support (-2LL=96.344, df=11, p < 0.001) Harasser-Victim -- 0.532 0.902† 0.763† Victim 0.079† -- *** 0.829† Harasser 1.225 1.121 -- 0.977 Uninvolved 1.428 0.949 0.608 -- Family Support (-2LL=75.392, df=10, p < 0.001) Harasser-Victim -- 1.004 159.493† 1.537† Victim 0.766† -- *** 0.310† Harasser 1.474 1.1354 -- 1.520 Uninvolved 0.578 1.039 *** -- Note. †=cell with less than 10 participants, *** indicates that the coefficient was not estimated 102 FIGURES Figure 3.1. Line graph showing relative error by the number of splits 103 Figure 3.2. Final DTA model for attrition HS School 1<.5 446/197 MS School 1<.5 557/473 Age<16 530/364 1 27/109 1 13/42 1 84/167 0 433/155 104 Figure 4.1. LCA probabilities at wave 1 LCA models with 3–5 classes for each wave. 0 0.2 0.4 0.6 0.8 1 1.2 1 2 3 Victim 1 Victim 2 Harass 1 Harass2 Harass3 0 0.2 0.4 0.6 0.8 1 1.2 1 2 3 4 Victim 1 Victim 2 Harass 1 Harass2 Harass3 0 0.2 0.4 0.6 0.8 1 1.2 1 2 3 4 5 Victim 1 Victim 2 Harass 1 Harass2 Harass3
Abstract (if available)
Abstract
Research has found that online victimization and harassment are related to negative outcomes for both victims and harassers. Further, a group involved in both victimization and harassment, harasser-victims, has the most detrimental outcomes. Using risk and resilience and social cognitive theories, I examined the risk for, development of, and factors related to cessation of online harassment and/or victimization over two waves of data. I employed latent transition analysis to examine underlying or latent groups, called statues, in the data. Latent transition analysis also allowed me to examine how adolescents changes status membership over time. By adding in covariates I investigated the relationship between social support from peers, family, and teachers and both initial statuses as well as transitions between statuses. Results indicated that offline peer, family, and teacher support were each related to higher odds of membership in the uninvolved class than the harasser-only, victim-only, or harasser-victim classes. Online peer support, however, was related to lower odds of membership in the uninvolved status than any of the other three statuses. Examination of the relationship between covariates and the transitions indicated that higher levels of offline peer, online peer, and family support were protective-stabilizing for some participants in the harasser-only status at wave one. However, all four types of social support were also related to greater odds of transitioning to the harasser-victim and victim-only statuses for other participants in the harasser-only status. Further, higher levels of teacher support and online peer support were related to higher odds of uninvolved participants transitioning to the victim-only status. Whereas family and offline peer support were related to lower odds of transitioning from uninvolved to the harasser-victim status at wave two. These results seem confounding, but the differences may be due to the type or quality of support an adolescent received. Though the model is exploratory in nature, the results indicate that schools and teachers as well friends and family members may have some influence over the changes in online harassment and victimization for adolescents.
Linked assets
University of Southern California Dissertations and Theses
Asset Metadata
Creator
Hiss, Sophia Caitlin
(author)
Core Title
An examination of in-school and online protective factors for adolescent trajectories of online victimization and harassment over time
School
Rossier School of Education
Degree
Doctor of Philosophy
Degree Program
Urban Education Policy
Publication Date
10/14/2015
Defense Date
08/24/2015
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
adolescence,cyberbullying,family support,OAI-PMH Harvest,online victimization,peer support,protective factors,social support,teacher support
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Tynes, Brendesha M. (
committee chair
), McArdle, John J. (
committee member
), Rueda, Robert (
committee member
)
Creator Email
hiss@usc.edu,sophie.hiss@gmail.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c40-190607
Unique identifier
UC11278125
Identifier
etd-HissSophia-3975.pdf (filename),usctheses-c40-190607 (legacy record id)
Legacy Identifier
etd-HissSophia-3975-1.pdf
Dmrecord
190607
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Hiss, Sophia Caitlin
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
cyberbullying
family support
online victimization
peer support
protective factors
social support
teacher support